Datasets:

---

JetBrains-Research

/

commit-chronicle

like 9

Follow

JetBrains Research 71

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/InstructionUtils.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/InstructionUtils.java", "diff": "@@ -1089,7 +1089,7 @@ public class InstructionUtils\nreturn InstructionUtils.concatOperands(parts[0], parts[1], createOperand(op1), createOperand(op2), createOperand(out));\n}\n- public static String constructUnaryInstString(String instString, CPOperand op1, String opcode, CPOperand out) {\n+ public static String constructUnaryInstString(String instString, String opcode, CPOperand op1, CPOperand out) {\nString[] parts = instString.split(Lop.OPERAND_DELIMITOR);\nparts[1] = opcode;\nreturn InstructionUtils.concatOperands(parts[0], parts[1], createOperand(op1), createOperand(out));\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/FEDInstructionUtils.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/FEDInstructionUtils.java", "diff": "package org.apache.sysds.runtime.instructions.fed;\nimport org.apache.commons.lang3.ArrayUtils;\n+\nimport org.apache.sysds.runtime.codegen.SpoofCellwise;\nimport org.apache.sysds.runtime.codegen.SpoofMultiAggregate;\nimport org.apache.sysds.runtime.codegen.SpoofOuterProduct;\n@@ -105,13 +106,14 @@ public class FEDInstructionUtils {\nelse if( inst instanceof MMChainCPInstruction) {\nMMChainCPInstruction linst = (MMChainCPInstruction) inst;\nMatrixObject mo = ec.getMatrixObject(linst.input1);\n- if( mo.isFederated() )\n+ if( mo.isFederated(FType.ROW) )\nfedinst = MMChainFEDInstruction.parseInstruction(linst.getInstructionString());\n}\nelse if( inst instanceof MMTSJCPInstruction ) {\nMMTSJCPInstruction linst = (MMTSJCPInstruction) inst;\nMatrixObject mo = ec.getMatrixObject(linst.input1);\n- if( mo.isFederated() )\n+ if( (mo.isFederated(FType.ROW) && linst.getMMTSJType().isLeft()) ||\n+ (mo.isFederated(FType.COL) && linst.getMMTSJType().isRight()))\nfedinst = TsmmFEDInstruction.parseInstruction(linst.getInstructionString());\n}\nelse if (inst instanceof UnaryCPInstruction && ! (inst instanceof IndexingCPInstruction)) {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/ParameterizedBuiltinFEDInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/ParameterizedBuiltinFEDInstruction.java", "diff": "@@ -179,7 +179,7 @@ public class ParameterizedBuiltinFEDInstruction extends ComputationFEDInstructio\nlong ncolId = FederationUtils.getNextFedDataID();\nCPOperand ncolOp = new CPOperand(String.valueOf(ncolId), ValueType.INT64, DataType.SCALAR);\n- String unaryString = InstructionUtils.constructUnaryInstString(instString, output, \"ncol\", ncolOp);\n+ String unaryString = InstructionUtils.constructUnaryInstString(instString, \"ncol\", ncolOp, output);\nFederatedRequest fr2 = FederationUtils.callInstruction(unaryString, ncolOp,\nnew CPOperand[] {output}, new long[] {out.getFedMapping().getID()});\nFederatedRequest fr3 = new FederatedRequest(FederatedRequest.RequestType.GET_VAR, fr2.getID());\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/TsmmFEDInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/TsmmFEDInstruction.java", "diff": "package org.apache.sysds.runtime.instructions.fed;\n+import java.util.concurrent.Future;\n+\nimport org.apache.sysds.lops.MMTSJ.MMTSJType;\nimport org.apache.sysds.runtime.DMLRuntimeException;\nimport org.apache.sysds.runtime.controlprogram.caching.MatrixObject;\nimport org.apache.sysds.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysds.runtime.controlprogram.federated.FederatedRequest;\n+import org.apache.sysds.runtime.controlprogram.federated.FederatedRequest.RequestType;\nimport org.apache.sysds.runtime.controlprogram.federated.FederatedResponse;\nimport org.apache.sysds.runtime.controlprogram.federated.FederationMap;\nimport org.apache.sysds.runtime.controlprogram.federated.FederationUtils;\n-import org.apache.sysds.runtime.controlprogram.federated.FederatedRequest.RequestType;\nimport org.apache.sysds.runtime.instructions.InstructionUtils;\nimport org.apache.sysds.runtime.instructions.cp.CPOperand;\nimport org.apache.sysds.runtime.matrix.data.MatrixBlock;\n-import java.util.concurrent.Future;\n-\npublic class TsmmFEDInstruction extends BinaryFEDInstruction {\nprivate final MMTSJType _type;\n@SuppressWarnings(\"unused\")\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/federated/algorithms/FederatedGLMTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/federated/algorithms/FederatedGLMTest.java", "diff": "package org.apache.sysds.test.functions.federated.algorithms;\n-import org.junit.Assert;\n-import org.junit.Test;\n-import org.junit.runner.RunWith;\n-import org.junit.runners.Parameterized;\n+import java.util.Arrays;\n+import java.util.Collection;\n+\nimport org.apache.sysds.common.Types;\nimport org.apache.sysds.common.Types.ExecMode;\nimport org.apache.sysds.runtime.meta.MatrixCharacteristics;\n@@ -30,9 +29,10 @@ import org.apache.sysds.runtime.util.HDFSTool;\nimport org.apache.sysds.test.AutomatedTestBase;\nimport org.apache.sysds.test.TestConfiguration;\nimport org.apache.sysds.test.TestUtils;\n-\n-import java.util.Arrays;\n-import java.util.Collection;\n+import org.junit.Assert;\n+import org.junit.Test;\n+import org.junit.runner.RunWith;\n+import org.junit.runners.Parameterized;\n@RunWith(value = Parameterized.class)\[email protected]\n@@ -47,6 +47,8 @@ public class FederatedGLMTest extends AutomatedTestBase {\npublic int rows;\[email protected](1)\npublic int cols;\n+ @Parameterized.Parameter(2)\n+ public boolean rowPartitioned;\n@Override\npublic void setUp() {\n@@ -58,8 +60,9 @@ public class FederatedGLMTest extends AutomatedTestBase {\npublic static Collection<Object[]> data() {\n// rows have to be even and > 1\nreturn Arrays.asList(new Object[][] {\n- // {10000, 10}, {1000, 100},\n- {2000, 43}});\n+ // {10000, 10, true}, {1000, 100, false},\n+ {2000, 44, true},\n+ {2000, 44, false}});\n}\n@Test\n@@ -79,16 +82,18 @@ public class FederatedGLMTest extends AutomatedTestBase {\nString HOME = SCRIPT_DIR + TEST_DIR;\n// write input matrices\n- int halfRows = rows / 2;\n+ int r = rowPartitioned ? rows / 2 : rows;\n+ int c = rowPartitioned ? cols : cols / 2;\n+\n// We have two matrices handled by a single federated worker\n- double[][] X1 = getRandomMatrix(halfRows, cols, 0, 1, 1, 42);\n- double[][] X2 = getRandomMatrix(halfRows, cols, 0, 1, 1, 1340);\n+ double[][] X1 = getRandomMatrix(r, c, 0, 1, 1, 42);\n+ double[][] X2 = getRandomMatrix(r, c, 0, 1, 1, 1340);\ndouble[][] Y = getRandomMatrix(rows, 1, -1, 1, 1, 1233);\nfor(int i = 0; i < rows; i++)\nY[i][0] = (Y[i][0] > 0) ? 1 : -1;\n- writeInputMatrixWithMTD(\"X1\", X1, false, new MatrixCharacteristics(halfRows, cols, blocksize, halfRows * cols));\n- writeInputMatrixWithMTD(\"X2\", X2, false, new MatrixCharacteristics(halfRows, cols, blocksize, halfRows * cols));\n+ writeInputMatrixWithMTD(\"X1\", X1, false, new MatrixCharacteristics(r, c, blocksize, r * c));\n+ writeInputMatrixWithMTD(\"X2\", X2, false, new MatrixCharacteristics(r, c, blocksize, r * c));\nwriteInputMatrixWithMTD(\"Y\", Y, false, new MatrixCharacteristics(rows, 1, blocksize, rows));\n// empty script name because we don't execute any script, just start the worker\n@@ -104,18 +109,18 @@ public class FederatedGLMTest extends AutomatedTestBase {\n// Run reference dml script with normal matrix\nfullDMLScriptName = HOME + TEST_NAME + \"Reference.dml\";\n- programArgs = new String[] {\"-args\", input(\"X1\"), input(\"X2\"), input(\"Y\"), expected(\"Z\")};\n+ programArgs = new String[] {\"-args\", input(\"X1\"), input(\"X2\"), input(\"Y\"), Boolean.toString(rowPartitioned).toUpperCase(), expected(\"Z\")};\nrunTest(true, false, null, -1);\n// Run actual dml script with federated matrix\nfullDMLScriptName = HOME + TEST_NAME + \".dml\";\nprogramArgs = new String[] {\"-stats\", \"-nvargs\", \"in_X1=\" + TestUtils.federatedAddress(port1, input(\"X1\")),\n\"in_X2=\" + TestUtils.federatedAddress(port2, input(\"X2\")), \"rows=\" + rows, \"cols=\" + cols,\n- \"in_Y=\" + input(\"Y\"), \"out=\" + output(\"Z\")};\n+ \"in_Y=\" + input(\"Y\"), \"rP=\" + Boolean.toString(rowPartitioned).toUpperCase(), \"out=\" + output(\"Z\")};\nrunTest(true, false, null, -1);\n// compare via files\n- compareResults(1e-9);\n+ compareResults(1e-2);\nTestUtils.shutdownThreads(t1, t2);\n@@ -124,7 +129,7 @@ public class FederatedGLMTest extends AutomatedTestBase {\nAssert.assertTrue(heavyHittersContainsString(\"fed_uark+\", \"fed_uarsqk+\"));\nAssert.assertTrue(heavyHittersContainsString(\"fed_uack+\"));\n// Assert.assertTrue(heavyHittersContainsString(\"fed_uak+\"));\n- Assert.assertTrue(heavyHittersContainsString(\"fed_mmchain\"));\n+ Assert.assertTrue(!rowPartitioned || heavyHittersContainsString(\"fed_mmchain\"));\n// check that federated input files are still existing\nAssert.assertTrue(HDFSTool.existsFileOnHDFS(input(\"X1\")));\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysds/test/functions/federated/primitives/FederatedMMChainTest.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysds.test.functions.federated.primitives;\n+\n+import java.util.Arrays;\n+import java.util.Collection;\n+\n+import org.apache.sysds.api.DMLScript;\n+import org.apache.sysds.common.Types.ExecMode;\n+import org.apache.sysds.runtime.meta.MatrixCharacteristics;\n+import org.apache.sysds.runtime.util.HDFSTool;\n+import org.apache.sysds.test.AutomatedTestBase;\n+import org.apache.sysds.test.TestConfiguration;\n+import org.apache.sysds.test.TestUtils;\n+import org.junit.Assert;\n+import org.junit.Test;\n+import org.junit.runner.RunWith;\n+import org.junit.runners.Parameterized;\n+\n+@RunWith(value = Parameterized.class)\[email protected]\n+public class FederatedMMChainTest extends AutomatedTestBase {\n+\n+ private final static String TEST_NAME1 = \"FederatedMMChainTest\";\n+ private final static String TEST_NAME2 = \"FederatedMMChainWeightsTest\";\n+ private final static String TEST_NAME3 = \"FederatedMMChainWeights2Test\";\n+\n+ private final static String TEST_DIR = \"functions/federated/\";\n+ private static final String TEST_CLASS_DIR = TEST_DIR + FederatedMMChainTest.class.getSimpleName() + \"/\";\n+\n+ private final static int blocksize = 1024;\n+ @Parameterized.Parameter()\n+ public int rows;\n+ @Parameterized.Parameter(1)\n+ public int cols;\n+ @Parameterized.Parameter(2)\n+ public boolean rowPartitioned;\n+\n+ @Parameterized.Parameters\n+ public static Collection<Object[]> data() {\n+ return Arrays.asList(new Object[][] {\n+ {1000, 100, true},\n+ {100, 1000, false}\n+ });\n+ }\n+\n+ @Override\n+ public void setUp() {\n+ TestUtils.clearAssertionInformation();\n+ addTestConfiguration(TEST_NAME1, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME1, new String[] {\"S\"}));\n+ addTestConfiguration(TEST_NAME2, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME2, new String[] {\"S\"}));\n+ addTestConfiguration(TEST_NAME3, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME3, new String[] {\"S\"}));\n+ }\n+\n+ @Test\n+ public void testMMChainCP() { runMMChainTest(ExecMode.SINGLE_NODE, TEST_NAME1); }\n+ @Test\n+ public void testMMChainWeightsCP() { runMMChainTest(ExecMode.SINGLE_NODE, TEST_NAME2); }\n+ @Test\n+ public void testMMChainWeights2CP() { runMMChainTest(ExecMode.SINGLE_NODE, TEST_NAME3); }\n+\n+ private void runMMChainTest(ExecMode execMode, String TEST_NAME) {\n+ boolean sparkConfigOld = DMLScript.USE_LOCAL_SPARK_CONFIG;\n+ ExecMode platformOld = rtplatform;\n+\n+ if(rtplatform == ExecMode.SPARK)\n+ DMLScript.USE_LOCAL_SPARK_CONFIG = true;\n+\n+ getAndLoadTestConfiguration(TEST_NAME);\n+ String HOME = SCRIPT_DIR + TEST_DIR;\n+\n+ // write input matrices\n+ int r = rows;\n+ int c = cols / 4;\n+ if(rowPartitioned) {\n+ r = rows / 4;\n+ c = cols;\n+ }\n+\n+ double[][] X1 = getRandomMatrix(r, c, 1, 5, 1, 3);\n+ double[][] X2 = getRandomMatrix(r, c, 1, 5, 1, 7);\n+ double[][] X3 = getRandomMatrix(r, c, 1, 5, 1, 8);\n+ double[][] X4 = getRandomMatrix(r, c, 1, 5, 1, 9);\n+\n+ MatrixCharacteristics mc = new MatrixCharacteristics(r, c, blocksize, r * c);\n+ writeInputMatrixWithMTD(\"X1\", X1, false, mc);\n+ writeInputMatrixWithMTD(\"X2\", X2, false, mc);\n+ writeInputMatrixWithMTD(\"X3\", X3, false, mc);\n+ writeInputMatrixWithMTD(\"X4\", X4, false, mc);\n+\n+ double[][] v = getRandomMatrix(cols, 1, 0, 1, 0.7, 3);\n+ writeInputMatrixWithMTD(\"v\", v, true);\n+ if(!TEST_NAME.equals(TEST_NAME1)){\n+ double[][] w = getRandomMatrix(rows, 1, 0, 1, 0.7, 10);\n+ writeInputMatrixWithMTD(\"w\", w, true);\n+ }\n+\n+ // empty script name because we don't execute any script, just start the worker\n+ fullDMLScriptName = \"\";\n+ int port1 = getRandomAvailablePort();\n+ int port2 = getRandomAvailablePort();\n+ int port3 = getRandomAvailablePort();\n+ int port4 = getRandomAvailablePort();\n+ Thread t1 = startLocalFedWorkerThread(port1, FED_WORKER_WAIT_S);\n+ Thread t2 = startLocalFedWorkerThread(port2, FED_WORKER_WAIT_S);\n+ Thread t3 = startLocalFedWorkerThread(port3, FED_WORKER_WAIT_S);\n+ Thread t4 = startLocalFedWorkerThread(port4);\n+\n+ rtplatform = execMode;\n+ if(rtplatform == ExecMode.SPARK) {\n+ DMLScript.USE_LOCAL_SPARK_CONFIG = true;\n+ }\n+ TestConfiguration config = availableTestConfigurations.get(TEST_NAME);\n+ loadTestConfiguration(config);\n+\n+ // Run reference dml script with normal matrix\n+ fullDMLScriptName = HOME + TEST_NAME + \"Reference.dml\";\n+ programArgs = new String[] {\"-stats\", \"100\", \"-args\", input(\"X1\"), input(\"X2\"), input(\"X3\"), input(\"X4\"),\n+ Boolean.toString(rowPartitioned).toUpperCase(), input(\"v\"), input(\"w\"), expected(\"S\")};\n+ runTest(null);\n+\n+ fullDMLScriptName = HOME + TEST_NAME + \".dml\";\n+ programArgs = new String[] {\"-stats\", \"100\", \"-nvargs\",\n+ \"in_X1=\" + TestUtils.federatedAddress(port1, input(\"X1\")),\n+ \"in_X2=\" + TestUtils.federatedAddress(port2, input(\"X2\")),\n+ \"in_X3=\" + TestUtils.federatedAddress(port3, input(\"X3\")),\n+ \"in_X4=\" + TestUtils.federatedAddress(port4, input(\"X4\")), \"rows=\" + rows, \"cols=\" + cols,\n+ \"rP=\" + Boolean.toString(rowPartitioned).toUpperCase(),\n+ \"in_v=\" + input(\"v\"),\n+ \"in_w=\" + input(\"w\"),\n+ \"out_S=\" + output(\"S\")};\n+ runTest(null);\n+\n+ // compare via files\n+ compareResults(1e-9);\n+\n+ // check that federated input files are still existing\n+ Assert.assertTrue(HDFSTool.existsFileOnHDFS(input(\"X1\")));\n+ Assert.assertTrue(HDFSTool.existsFileOnHDFS(input(\"X2\")));\n+ Assert.assertTrue(HDFSTool.existsFileOnHDFS(input(\"X3\")));\n+ Assert.assertTrue(HDFSTool.existsFileOnHDFS(input(\"X4\")));\n+\n+ TestUtils.shutdownThreads(t1, t2, t3, t4);\n+\n+ rtplatform = platformOld;\n+ DMLScript.USE_LOCAL_SPARK_CONFIG = sparkConfigOld;\n+ }\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/federated/FederatedGLMTest.dml", "new_path": "src/test/scripts/functions/federated/FederatedGLMTest.dml", "diff": "# under the License.\n#\n#-------------------------------------------------------------\n-\n+if ($rP) {\nX = federated(addresses=list($in_X1, $in_X2),\nranges=list(list(0, 0), list($rows / 2, $cols), list($rows / 2, 0), list($rows, $cols)))\n+} else {\n+ X = federated(addresses=list($in_X1, $in_X2),\n+ ranges=list(list(0, 0), list($rows, $cols / 2), list(0, $cols / 2), list($rows, $cols)))\n+}\nY = read($in_Y)\nmodel = glm(X=X, Y=Y, icpt = FALSE, tol = 1e-6, reg = 0.01)\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/federated/FederatedGLMTestReference.dml", "new_path": "src/test/scripts/functions/federated/FederatedGLMTestReference.dml", "diff": "#\n#-------------------------------------------------------------\n+if ($4) {\nX = rbind(read($1), read($2))\n+} else {\n+ X = cbind(read($1), read($2))\n+}\nY = read($3)\nmodel = glm(X=X, Y=Y, icpt = FALSE, tol = 1e-6, reg = 0.01)\n-write(model, $4)\n+write(model, $5)\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/federated/FederatedMMChainTest.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+if ($rP) {\n+ X = federated(addresses=list($in_X1, $in_X2, $in_X3, $in_X4),\n+ ranges=list(list(0, 0), list($rows/4, $cols), list($rows/4, 0), list(2*$rows/4, $cols),\n+ list(2*$rows/4, 0), list(3*$rows/4, $cols), list(3*$rows/4, 0), list($rows, $cols)));\n+ } else {\n+ X = federated(addresses=list($in_X1, $in_X2, $in_X3, $in_X4),\n+ ranges=list(list(0, 0), list($rows, $cols/4), list(0,$cols/4), list($rows, $cols/2),\n+ list(0,$cols/2), list($rows, 3*($cols/4)), list(0, 3*($cols/4)), list($rows, $cols)));\n+ }\n+\n+v = read($in_v);\n+\n+S = (t(X) %*% (X %*% v));\n+print(nrow(S))\n+print(ncol(S))\n+write(S, $out_S);\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/federated/FederatedMMChainTestReference.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+if($5) { X = rbind(read($1), read($2), read($3), read($4)); }\n+else { X = cbind(read($1), read($2), read($3), read($4));}\n+\n+v = read($6);\n+\n+S = (t(X) %*% (X %*% v));\n+\n+write(S, $8);\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/federated/FederatedMMChainWeights2Test.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+if ($rP) {\n+ X = federated(addresses=list($in_X1, $in_X2, $in_X3, $in_X4),\n+ ranges=list(list(0, 0), list($rows/4, $cols), list($rows/4, 0), list(2*$rows/4, $cols),\n+ list(2*$rows/4, 0), list(3*$rows/4, $cols), list(3*$rows/4, 0), list($rows, $cols)));\n+ } else {\n+ X = federated(addresses=list($in_X1, $in_X2, $in_X3, $in_X4),\n+ ranges=list(list(0, 0), list($rows, $cols/4), list(0,$cols/4), list($rows, $cols/2),\n+ list(0,$cols/2), list($rows, 3*($cols/4)), list(0, 3*($cols/4)), list($rows, $cols)));\n+ }\n+\n+v = read($in_v);\n+w = read($in_w);\n+S = t(X) %*% ((X %*% v)-w);\n+\n+write(S, $out_S);\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/federated/FederatedMMChainWeights2TestReference.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+if($5) { X = rbind(read($1), read($2), read($3), read($4)); }\n+else { X = cbind(read($1), read($2), read($3), read($4));}\n+\n+v = read($6);\n+w = read($7);\n+S = t(X) %*% ((X %*% v)-w);\n+\n+write(S, $8);\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/federated/FederatedMMChainWeightsTest.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+if ($rP) {\n+ X = federated(addresses=list($in_X1, $in_X2, $in_X3, $in_X4),\n+ ranges=list(list(0, 0), list($rows/4, $cols), list($rows/4, 0), list(2*$rows/4, $cols),\n+ list(2*$rows/4, 0), list(3*$rows/4, $cols), list(3*$rows/4, 0), list($rows, $cols)));\n+ } else {\n+ X = federated(addresses=list($in_X1, $in_X2, $in_X3, $in_X4),\n+ ranges=list(list(0, 0), list($rows, $cols/4), list(0,$cols/4), list($rows, $cols/2),\n+ list(0,$cols/2), list($rows, 3*($cols/4)), list(0, 3*($cols/4)), list($rows, $cols)));\n+ }\n+\n+v = read($in_v);\n+w = read($in_w);\n+S = (t(X) %*% (w*(X %*% v)));\n+\n+write(S, $out_S);\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/federated/FederatedMMChainWeightsTestReference.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+if($5) { X = rbind(read($1), read($2), read($3), read($4)); }\n+else { X = cbind(read($1), read($2), read($3), read($4));}\n+\n+v = read($6);\n+w = read($7);\n+S = (t(X) %*% (w*(X %*% v)));\n+\n+write(S, $8);\n" } ]

[ { "change_type": "MODIFY", "old_path": "NOTICE", "new_path": "NOTICE", "diff": "Apache SystemDS\n-Copyright [2015-2020] The Apache Software Foundation\n+Copyright [2015-2021] The Apache Software Foundation\nThis product includes software developed at\nThe Apache Software Foundation (http://www.apache.org/).\n" }, { "change_type": "MODIFY", "old_path": "src/assembly/bin/NOTICE", "new_path": "src/assembly/bin/NOTICE", "diff": "Apache SystemDS\n-Copyright [2015-2020] The Apache Software Foundation\n+Copyright [2015-2021] The Apache Software Foundation\nThis product includes software developed at\nThe Apache Software Foundation (http://www.apache.org/).\n" }, { "change_type": "MODIFY", "old_path": "src/assembly/extra/NOTICE", "new_path": "src/assembly/extra/NOTICE", "diff": "Apache SystemDS\n-Copyright [2015-2020] The Apache Software Foundation\n+Copyright [2015-2021] The Apache Software Foundation\nThis product includes software developed at\nThe Apache Software Foundation (http://www.apache.org/).\n" }, { "change_type": "MODIFY", "old_path": "src/main/python/docs/source/conf.py", "new_path": "src/main/python/docs/source/conf.py", "diff": "@@ -34,7 +34,7 @@ sys.path.insert(0, os.path.abspath('../..'))\n# -- Project information -----------------------------------------------------\nproject = 'SystemDS'\n-copyright = '2020, Apache SystemDS'\n+copyright = '2021, Apache SystemDS'\nauthor = 'Apache SystemDS'\n# The full version, including alpha/beta/rc tags\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/ParameterizedBuiltinFEDInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/ParameterizedBuiltinFEDInstruction.java", "diff": "@@ -179,7 +179,7 @@ public class ParameterizedBuiltinFEDInstruction extends ComputationFEDInstructio\nlong ncolId = FederationUtils.getNextFedDataID();\nCPOperand ncolOp = new CPOperand(String.valueOf(ncolId), ValueType.INT64, DataType.SCALAR);\n- String unaryString = InstructionUtils.constructUnaryInstString(instString, \"ncol\", ncolOp, output);\n+ String unaryString = InstructionUtils.constructUnaryInstString(instString, \"ncol\", output, ncolOp);\nFederatedRequest fr2 = FederationUtils.callInstruction(unaryString, ncolOp,\nnew CPOperand[] {output}, new long[] {out.getFedMapping().getID()});\nFederatedRequest fr3 = new FederatedRequest(FederatedRequest.RequestType.GET_VAR, fr2.getID());\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/io/ListReader.java", "new_path": "src/main/java/org/apache/sysds/runtime/io/ListReader.java", "diff": "@@ -51,7 +51,7 @@ public class ListReader\n* @param fmtStr format string\n* @param props file format properties\n* @return list object\n- * @throws DMLRuntimeException\n+ * @throws DMLRuntimeException if inconsistent meta data or read fails\n*/\npublic static ListObject readListFromHDFS(String fname, String fmtStr, FileFormatProperties props)\nthrows DMLRuntimeException\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/io/ListWriter.java", "new_path": "src/main/java/org/apache/sysds/runtime/io/ListWriter.java", "diff": "@@ -43,7 +43,7 @@ public class ListWriter\n* @param fname directory name\n* @param fmtStr format string\n* @param props file format properties\n- * @throws DMLRuntimeException\n+ * @throws DMLRuntimeException if write fails\n*/\npublic static void writeListToHDFS(ListObject lo, String fname, String fmtStr, FileFormatProperties props)\nthrows DMLRuntimeException\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/util/HDFSTool.java", "new_path": "src/main/java/org/apache/sysds/runtime/util/HDFSTool.java", "diff": "@@ -313,7 +313,15 @@ public class HDFSTool\npublic static ScalarObject readScalarObjectFromHDFSFile(String fname, ValueType vt) {\ntry {\n- return ScalarObjectFactory.createScalarObject(vt, readObjectFromHDFSFile(fname, vt));\n+ Object obj = null;\n+ switch( vt ) {\n+ case INT64: obj = readIntegerFromHDFSFile(fname); break;\n+ case FP64: obj = readDoubleFromHDFSFile(fname); break;\n+ case BOOLEAN: obj = readBooleanFromHDFSFile(fname); break;\n+ case STRING:\n+ default: obj = readStringFromHDFSFile(fname);\n+ }\n+ return ScalarObjectFactory.createScalarObject(vt, obj);\n}\ncatch(Exception ex) {\nthrow new DMLRuntimeException(ex);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/util/ProgramConverter.java", "new_path": "src/main/java/org/apache/sysds/runtime/util/ProgramConverter.java", "diff": "@@ -1120,14 +1120,15 @@ public class ProgramConverter\nfor( String fkey : prog.getFunctionProgramBlocks().keySet() ) {\nif( !cand.contains(fkey) ) //skip function not included in the parfor body\ncontinue;\n- if( count>0 ) {\n+ if( count>0 )\nsb.append( ELEMENT_DELIM );\n- }\nsb.append( fkey );\nsb.append( KEY_VALUE_DELIM );\nFunctionProgramBlock fpb1 = prog.getFunctionProgramBlock(fkey, true);\nsb.append( rSerializeProgramBlock(fpb1, clsMap) );\nif( prog.containsFunctionProgramBlock(fkey, false) ) {\n+ sb.append( ELEMENT_DELIM );\n+ sb.append( fkey );\nsb.append( KEY_VALUE_DELIM );\nFunctionProgramBlock fpb2 = prog.getFunctionProgramBlock(fkey, false);\nsb.append( rSerializeProgramBlock(fpb2, clsMap) );\n@@ -1392,21 +1393,11 @@ public class ProgramConverter\nString lvar = st.nextToken(); //with ID = CP_CHILD_THREAD+id for current use\n//put first copy into prog (for direct use)\nint index = lvar.indexOf( KEY_VALUE_DELIM );\n- String tmp1 = lvar.substring(0, index);\n- String tmp2 = lvar.substring(index + 1);\n- if( tmp2.contains(KEY_VALUE_DELIM) ) {\n- int index2 = tmp2.indexOf( KEY_VALUE_DELIM );\n- String tmp21 = tmp2.substring(0, index2);\n- String tmp22 = tmp2.substring(index2 + 1);\n- prog.addFunctionProgramBlock(tmp1,\n- (FunctionProgramBlock)rParseProgramBlock(tmp21, prog, id), true);\n- prog.addFunctionProgramBlock(tmp1,\n- (FunctionProgramBlock)rParseProgramBlock(tmp22, prog, id), false);\n- }\n- else {\n- prog.addFunctionProgramBlock(tmp1,\n- (FunctionProgramBlock)rParseProgramBlock(tmp2, prog, id), true);\n- }\n+ String fkey = lvar.substring(0, index);\n+ String tmp = lvar.substring(index + 1);\n+ boolean opt = !prog.containsFunctionProgramBlock(fkey, true);\n+ prog.addFunctionProgramBlock(fkey,\n+ (FunctionProgramBlock)rParseProgramBlock(tmp, prog, id), opt);\n}\nreturn ret;\n}\n" } ]

[ { "change_type": "DELETE", "old_path": "notebooks/databricks/MLContext.scala", "new_path": null, "diff": "-// Databricks notebook source\n-/*\n- * Licensed to the Apache Software Foundation (ASF) under one\n- * or more contributor license agreements. See the NOTICE file\n- * distributed with this work for additional information\n- * regarding copyright ownership. The ASF licenses this file\n- * to you under the Apache License, Version 2.0 (the\n- * \"License\"); you may not use this file except in compliance\n- * with the License. You may obtain a copy of the License at\n- *\n- * http://www.apache.org/licenses/LICENSE-2.0\n- *\n- * Unless required by applicable law or agreed to in writing,\n- * software distributed under the License is distributed on an\n- * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n- * KIND, either express or implied. See the License for the\n- * specific language governing permissions and limitations\n- * under the License.\n- */\n-\n-\n-// COMMAND ----------\n-\n-// MAGIC %md # Apache SystemDS on Databricks\n-\n-// COMMAND ----------\n-\n-// MAGIC %md ## Create a quickstart cluster\n-// MAGIC\n-// MAGIC 1. In the sidebar, right-click the **Clusters** button and open the link in a new window.\n-// MAGIC 1. On the Clusters page, click **Create Cluster**.\n-// MAGIC 1. Name the cluster **Quickstart**.\n-// MAGIC 1. In the Databricks Runtime Version drop-down, select **6.4 (Scala 2.11, Spark 2.4.5)**.\n-// MAGIC 1. Click **Create Cluster**.\n-// MAGIC 1. Attach `SystemDS.jar` file to the libraries\n-\n-// COMMAND ----------\n-\n-// MAGIC %md ## Attach the notebook to the cluster and run all commands in the notebook\n-// MAGIC\n-// MAGIC 1. Return to this notebook.\n-// MAGIC 1. In the notebook menu bar, select **<img src=\"http://docs.databricks.com/_static/images/notebooks/detached.png\"/></a> > Quickstart**.\n-// MAGIC 1. When the cluster changes from <img src=\"http://docs.databricks.com/_static/images/clusters/cluster-starting.png\"/></a> to <img src=\"http://docs.databricks.com/_static/images/clusters/cluster-running.png\"/></a>, click **<img src=\"http://docs.databricks.com/_static/images/notebooks/run-all.png\"/></a> Run All**.\n-\n-// COMMAND ----------\n-\n-// MAGIC %md ## Load SystemDS MLContext API\n-\n-// COMMAND ----------\n-\n-import org.apache.sysds.api.mlcontext._\n-import org.apache.sysds.api.mlcontext.ScriptFactory._\n-val ml = new MLContext(spark)\n-\n-// COMMAND ----------\n-\n-val habermanUrl = \"http://archive.ics.uci.edu/ml/machine-learning-databases/haberman/haberman.data\"\n-val habermanList = scala.io.Source.fromURL(habermanUrl).mkString.split(\"\\n\")\n-val habermanRDD = sc.parallelize(habermanList)\n-val habermanMetadata = new MatrixMetadata(306, 4)\n-val typesRDD = sc.parallelize(Array(\"1.0,1.0,1.0,2.0\"))\n-val typesMetadata = new MatrixMetadata(1, 4)\n-val scriptUrl = \"https://raw.githubusercontent.com/apache/systemds/master/scripts/algorithms/Univar-Stats.dml\"\n-val uni = dmlFromUrl(scriptUrl).in(\"A\", habermanRDD, habermanMetadata).in(\"K\", typesRDD, typesMetadata).in(\"$CONSOLE_OUTPUT\", true)\n-ml.execute(uni)\n-\n-// COMMAND ----------\n-\n-// MAGIC %md ### Create a neural network layer with (R-like) DML language\n-\n-// COMMAND ----------\n-\n-val s = \"\"\"\n- source(\"scripts/nn/layers/relu.dml\") as relu;\n- X = rand(rows=100, cols=10, min=-1, max=1);\n- R1 = relu::forward(X);\n- R2 = max(X, 0);\n- R = sum(R1==R2);\n- \"\"\"\n-\n-val ret = ml.execute(dml(s).out(\"R\")).getScalarObject(\"R\").getDoubleValue();\n-\n-// COMMAND ----------\n-\n-// MAGIC %md ### Recommendation with Amazon review dataset\n-\n-// COMMAND ----------\n-\n-import java.net.URL\n-import java.io.File\n-import org.apache.commons.io.FileUtils\n-\n-FileUtils.copyURLToFile(new URL(\"http://snap.stanford.edu/data/amazon0601.txt.gz\"), new File(\"/tmp/amazon0601.txt.gz\"))\n-\n-// COMMAND ----------\n-\n-// MAGIC %sh\n-// MAGIC gunzip -d /tmp/amazon0601.txt.gz\n-\n-// COMMAND ----------\n-\n-// To list the file system files. For more https://docs.databricks.com/data/filestore.html\n-// File system: display(dbutils.fs.ls(\"file:/tmp\"))\n-// DBFS: display(dbutils.fs.ls(\".\"))\n-\n-dbutils.fs.mv(\"file:/tmp/amazon0601.txt\", \"dbfs:/tmp/amazon0601.txt\")\n-\n-// COMMAND ----------\n-\n-display(dbutils.fs.ls(\"/tmp\"))\n-// display(dbutils.fs.ls(\"file:/tmp\"))\n-\n-// COMMAND ----------\n-\n-// move temporary files to databricks file system (DBFS)\n-// dbutils.fs.mv(\"file:/databricks/driver/amazon0601.txt\", \"dbfs:/tmp/amazon0601.txt\")\n-val df = spark.read.format(\"text\").option(\"inferSchema\", \"true\").option(\"header\",\"true\").load(\"dbfs:/tmp/amazon0601.txt\")\n-display(df)\n-\n-// COMMAND ----------\n-\n-// MAGIC %py\n-// MAGIC\n-// MAGIC # The scala data processing pipeline can also be\n-// MAGIC # implemented in python as shown in this block\n-// MAGIC\n-// MAGIC #\n-// MAGIC # import pyspark.sql.functions as F\n-// MAGIC # # https://spark.apache.org/docs/latest/sql-ref.html\n-// MAGIC\n-// MAGIC # dataPath = \"dbfs:/tmp/amazon0601.txt\"\n-// MAGIC\n-// MAGIC # X_train = (sc.textFile(dataPath)\n-// MAGIC # .filter(lambda l: not l.startswith(\"#\"))\n-// MAGIC # .map(lambda l: l.split(\"\\t\"))\n-// MAGIC # .map(lambda prods: (int(prods[0]), int(prods[1]), 1.0))\n-// MAGIC # .toDF((\"prod_i\", \"prod_j\", \"x_ij\"))\n-// MAGIC # .filter(\"prod_i < 500 AND prod_j < 500\") # Filter for memory constraints\n-// MAGIC # .cache())\n-// MAGIC\n-// MAGIC # max_prod_i = X_train.select(F.max(\"prod_i\")).first()[0]\n-// MAGIC # max_prod_j = X_train.select(F.max(\"prod_j\")).first()[0]\n-// MAGIC # numProducts = max(max_prod_i, max_prod_j) + 1 # 0-based indexing\n-// MAGIC # print(\"Total number of products: {}\".format(numProducts))\n-\n-// COMMAND ----------\n-\n-// Reference: https://spark.apache.org/docs/latest/rdd-programming-guide.html\n-val X_train = (sc.textFile(\"dbfs:/tmp/amazon0601.txt\").filter(l => !(l.startsWith(\"#\"))).map(l => l.split(\"\\t\"))\n- .map(prods => (prods(0).toLong, prods(1).toLong, 1.0))\n- .toDF(\"prod_i\", \"prod_j\", \"x_ij\")\n- .filter(\"prod_i < 500 AND prod_j < 500\") // filter for memory constraints\n- .cache())\n-\n-display(X_train)\n-\n-// COMMAND ----------\n-\n-// MAGIC %md #### Poisson Nonnegative Matrix Factorization\n-\n-// COMMAND ----------\n-\n-# Poisson Nonnegative Matrix Factorization\n-\n-val pnmf = \"\"\"\n-# data & args\n-X = X+1 # change product IDs to be 1-based, rather than 0-based\n-V = table(X[,1], X[,2])\n-size = ifdef($size, -1)\n-if(size > -1) {\n- V = V[1:size,1:size]\n-}\n-\n-n = nrow(V)\n-m = ncol(V)\n-range = 0.01\n-W = Rand(rows=n, cols=rank, min=0, max=range, pdf=\"uniform\")\n-H = Rand(rows=rank, cols=m, min=0, max=range, pdf=\"uniform\")\n-losses = matrix(0, rows=max_iter, cols=1)\n-\n-# run PNMF\n-i=1\n-while(i <= max_iter) {\n- # update params\n- H = (H * (t(W) %*% (V/(W%*%H))))/t(colSums(W))\n- W = (W * ((V/(W%*%H)) %*% t(H)))/t(rowSums(H))\n-\n- # compute loss\n- losses[i,] = -1 * (sum(V*log(W%*%H)) - as.scalar(colSums(W)%*%rowSums(H)))\n- i = i + 1;\n-}\n- \"\"\"\n-\n-val ret = ml.execute(dml(pnmf).in(\"X\", X_train).in(\"max_iter\", 100).in(\"rank\", 10).out(\"W\").out(\"H\").out(\"losses\"));\n-\n-// COMMAND ----------\n-\n-val W = ret.getMatrix(\"W\")\n-val H = ret.getMatrix(\"H\")\n-val losses = ret.getMatrix(\"losses\")\n-\n-// COMMAND ----------\n-\n-val lossesDF = losses.toDF().sort(\"__INDEX\")\n-display(lossesDF)\n" }, { "change_type": "DELETE", "old_path": "notebooks/databricks/README.md", "new_path": null, "diff": "-#### Setup Apache SystemDS on Databricks platform\n-\n-1. Create a new account at [databricks cloud](https://community.cloud.databricks.com/)\n-2. In left-side navbar select **Clusters** > **`+ Create Cluster`** > Name the cluster! > **`Create Cluster`**\n-3. Navigate to the created cluster configuration.\n- 1. Select **Libraries**\n- 2. Select **Install New** > **Library Source [`Upload`]** and **Library Type [`Jar`]**\n- 3. Upload the `SystemDS.jar` file! > **`Install`**\n-4. Attach a notebook to the cluster above.\n" } ]

[ { "change_type": "MODIFY", "old_path": "docs/_includes/scripts.html", "new_path": "docs/_includes/scripts.html", "diff": "@@ -16,24 +16,7 @@ See the License for the specific language governing permissions and\nlimitations under the License.\n{% endcomment %}\n-->\n-{% if site.analytics_on == true %} {% case site.analytics_provider %} {% when \"google_universal\" %}\n-<!-- Analytics -->\n-<script>\n- (function(i, s, o, g, r, a, m) {\n- i['GoogleAnalyticsObject'] = r;\n- i[r] = i[r] || function() {\n- (i[r].q = i[r].q || []).push(arguments)\n- }, i[r].l = 1 * new Date();\n- a = s.createElement(o),\n- m = s.getElementsByTagName(o)[0];\n- a.async = 1;\n- a.src = g;\n- m.parentNode.insertBefore(a, m)\n- })(window, document, 'script', '//www.google-analytics.com/analytics.js', 'ga');\n- ga('create', '{{ site.analytics_google_universal_tracking_id }}', 'auto');\n- ga('send', 'pageview');\n-</script>\n-{% endcase %} {% endif %}\n+\n<!-- MathJax Section -->\n<script type=\"text/x-mathjax-config\">\n@@ -67,19 +50,3 @@ limitations under the License.\nd.getElementsByTagName('head')[0].appendChild(script);\n}(document));\n</script>\n-<!-- Algolia search section -->\n-<script type=\"text/javascript\" src=\"https://cdn.jsdelivr.net/npm/docsearch.js@2/dist/cdn/docsearch.min.js\"></script>\n-<script>\n- // Crawler configuration for the search indexing is available at:\n- // https://github.com/algolia/docsearch-configs/blob/master/configs/apache_systemml.json\n-\n- docsearch({\n- apiKey: '78c19564c220d4642a41197baae304ef',\n- indexName: 'apache_systemds',\n- inputSelector: \"#s-bar\",\n- // For custom styling for the dropdown, please set debug to true\n- // so that the dropdown won't disappear when the inspect tools are\n- // open.\n- debug: false\n- });\n-</script>\n" }, { "change_type": "MODIFY", "old_path": "docs/_layouts/base.html", "new_path": "docs/_layouts/base.html", "diff": "@@ -34,7 +34,6 @@ limitations under the License.\n<link rel=\"stylesheet\" href=\"./css/main.css\">\n<link rel=\"stylesheet\" href=\"./css/pygments-default.css\">\n<link rel=\"shortcut icon\" href=\"./img/favicon.png\">\n- <!-- <link rel=\"stylesheet\" href=\"https://cdn.jsdelivr.net/npm/docsearch.js@2/dist/cdn/docsearch.min.css\" /> -->\n<script src=\"./js/vendor/jquery-1.12.0.min.js\"></script>\n<script src=\"./js/vendor/bootstrap.min.js\"></script>\n<script src=\"./js/vendor/anchor.min.js\"></script>\n" }, { "change_type": "MODIFY", "old_path": "docs/_layouts/site.html", "new_path": "docs/_layouts/site.html", "diff": "@@ -34,7 +34,6 @@ limitations under the License.\n<link rel=\"stylesheet\" href=\"./../css/main.css\">\n<link rel=\"stylesheet\" href=\"./../css/pygments-default.css\">\n<link rel=\"shortcut icon\" href=\"./../img/favicon.png\">\n- <link rel=\"stylesheet\" href=\"https://cdn.jsdelivr.net/npm/docsearch.js@2/dist/cdn/docsearch.min.css\" />\n<script src=\"./../js/vendor/jquery-1.12.0.min.js\"></script>\n<script src=\"./../js/vendor/bootstrap.min.js\"></script>\n<script src=\"./../js/vendor/anchor.min.js\"></script>\n" }, { "change_type": "MODIFY", "old_path": "docs/css/main.css", "new_path": "docs/css/main.css", "diff": "@@ -336,7 +336,6 @@ table {\n.dropdown-menu li>a:hover {\nbackground-color: #ff5003;\nbackground-image: none;\n- font-color: red;\n}\na {\n" }, { "change_type": "MODIFY", "old_path": "docs/css/pygments-default.css", "new_path": "docs/css/pygments-default.css", "diff": "@@ -15,7 +15,6 @@ insert the code (or pre?) tags for you.\n.hll { background-color: #ffffcc }\n.c { color: #60a0b0; font-style: italic } /* Comment */\n-.err { } /* Error */\n.k { color: #007020; font-weight: bold } /* Keyword */\n.o { color: #666666 } /* Operator */\n.cm { color: #60a0b0; font-style: italic } /* Comment.Multiline */\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/CompressedMatrixBlockFactory.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/CompressedMatrixBlockFactory.java", "diff": "@@ -170,7 +170,6 @@ public class CompressedMatrixBlockFactory {\n}\nprivate void coCodePhase(CompressedSizeEstimator sizeEstimator, CompressedSizeInfo sizeInfos, int numRows) {\n- // for(int i = 0; i < 100000; i ++)\ncoCodeColGroups = PlanningCoCoder.findCoCodesByPartitioning(sizeEstimator, sizeInfos, numRows, k, compSettings);\n_stats.estimatedSizeCoCoded = coCodeColGroups.memoryEstimate();\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/cocode/CoCodeBinPacking.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/cocode/CoCodeBinPacking.java", "diff": "@@ -22,12 +22,16 @@ package org.apache.sysds.runtime.compress.cocode;\nimport java.util.ArrayList;\nimport java.util.Arrays;\nimport java.util.Comparator;\n+import java.util.HashMap;\nimport java.util.List;\n+import java.util.Map;\nimport org.apache.sysds.runtime.compress.CompressionSettings;\n+import org.apache.sysds.runtime.compress.colgroup.AColGroup.CompressionType;\nimport org.apache.sysds.runtime.compress.estim.CompressedSizeEstimator;\nimport org.apache.sysds.runtime.compress.estim.CompressedSizeInfo;\nimport org.apache.sysds.runtime.compress.estim.CompressedSizeInfoColGroup;\n+import org.apache.sysds.runtime.compress.utils.Util;\n/**\n* Column group partitioning with bin packing heuristic.\n@@ -38,6 +42,8 @@ public class CoCodeBinPacking extends AColumnCoCoder {\nprivate static final int MAX_COL_FIRST_FIT = 16384;\nprivate static final int MAX_COL_PER_GROUP = 1024;\n+ private final Memorizer mem;\n+\n/**\n* Use a constant partition size independent of the number of rows in order to ensure constant compression speed\n* independent of blocking. Higher values gives more CoCoding at the cost of longer compressionTimes.\n@@ -48,11 +54,32 @@ public class CoCodeBinPacking extends AColumnCoCoder {\nprotected CoCodeBinPacking(CompressedSizeEstimator sizeEstimator, CompressionSettings cs) {\nsuper(sizeEstimator, cs);\n+ mem = new Memorizer();\n}\n@Override\nprotected CompressedSizeInfo coCodeColumns(CompressedSizeInfo colInfos, int k) {\n- colInfos.setInfo(partitionColumns(colInfos.getInfo()));\n+ // establish memo table for extracted column groups\n+\n+ List<CompressedSizeInfoColGroup> constantGroups = new ArrayList<>();\n+ List<CompressedSizeInfoColGroup> newGroups = new ArrayList<>();\n+\n+ for(CompressedSizeInfoColGroup g : colInfos.getInfo()) {\n+ if(g.getBestCompressionType(_cs) == CompressionType.CONST)\n+ constantGroups.add(g);\n+ else {\n+ mem.put(g);\n+ newGroups.add(g);\n+ }\n+ }\n+\n+ // make bins\n+ colInfos.setInfo(partitionColumns(newGroups));\n+ // Cocode compare all in bins\n+ getCoCodingGroupsBruteForce(colInfos, k);\n+\n+ colInfos.getInfo().addAll(constantGroups);\n+\nreturn colInfos;\n}\n@@ -130,4 +157,186 @@ public class CoCodeBinPacking extends AColumnCoCoder {\nreturn bins;\n}\n+\n+ /**\n+ * This methods verifies the coCoded bins actually are the best combinations within each individual Group based on\n+ * the sample.\n+ *\n+ * @param bins The bins constructed based on lightweight estimations\n+ * @param k The number of threads allowed to be used.\n+ * @param est The Estimator to be used.\n+ * @return\n+ */\n+ private CompressedSizeInfo getCoCodingGroupsBruteForce(CompressedSizeInfo bins, int k) {\n+\n+ List<CompressedSizeInfoColGroup> finalGroups = new ArrayList<>();\n+ // For each bin of columns that is allowed to potentially cocode.\n+ for(CompressedSizeInfoColGroup bin : bins.getInfo()) {\n+ final int len = bin.getColumns().length;\n+ if(len == 0)\n+ continue;\n+ else if(len == 1)\n+ // early termination\n+ finalGroups.add(bin);\n+ else\n+ finalGroups.addAll(coCodeBruteForce(bin));\n+ }\n+\n+ bins.setInfo(finalGroups);\n+ return bins;\n+ }\n+\n+ private List<CompressedSizeInfoColGroup> coCodeBruteForce(CompressedSizeInfoColGroup bin) {\n+\n+ List<int[]> workset = new ArrayList<>(bin.getColumns().length);\n+\n+ for(int i = 0; i < bin.getColumns().length; i++)\n+ workset.add(new int[] {bin.getColumns()[i]});\n+\n+ // process merging iterations until no more change\n+ while(workset.size() > 1) {\n+ long changeInSize = 0;\n+ CompressedSizeInfoColGroup tmp = null;\n+ int[] selected1 = null, selected2 = null;\n+ for(int i = 0; i < workset.size(); i++) {\n+ for(int j = i + 1; j < workset.size(); j++) {\n+ final int[] c1 = workset.get(i);\n+ final int[] c2 = workset.get(j);\n+ final long sizeC1 = mem.get(c1).getMinSize();\n+ final long sizeC2 = mem.get(c2).getMinSize();\n+\n+ mem.incst1();\n+ // pruning filter : skip dominated candidates\n+ // Since even if the entire size of one of the column lists is removed,\n+ // it still does not improve compression\n+ if(-Math.min(sizeC1, sizeC2) > changeInSize)\n+ continue;\n+\n+ // Join the two column groups.\n+ // and Memorize the new join.\n+ final CompressedSizeInfoColGroup c1c2Inf = mem.getOrCreate(c1, c2);\n+ final long sizeC1C2 = c1c2Inf.getMinSize();\n+\n+ long newSizeChangeIfSelected = sizeC1C2 - sizeC1 - sizeC2;\n+ // Select the best join of either the currently selected\n+ // or keep the old one.\n+ if((tmp == null && newSizeChangeIfSelected < changeInSize) || tmp != null &&\n+ (newSizeChangeIfSelected < changeInSize || newSizeChangeIfSelected == changeInSize &&\n+ c1c2Inf.getColumns().length < tmp.getColumns().length)) {\n+ changeInSize = newSizeChangeIfSelected;\n+ tmp = c1c2Inf;\n+ selected1 = c1;\n+ selected2 = c2;\n+ }\n+ }\n+ }\n+\n+ if(tmp != null) {\n+ workset.remove(selected1);\n+ workset.remove(selected2);\n+ workset.add(tmp.getColumns());\n+ }\n+ else\n+ break;\n+ }\n+\n+ LOG.debug(mem.stats());\n+ mem.resetStats();\n+\n+ List<CompressedSizeInfoColGroup> ret = new ArrayList<>(workset.size());\n+\n+ for(int[] w : workset)\n+ ret.add(mem.get(w));\n+\n+ return ret;\n+ }\n+\n+ protected class Memorizer {\n+ private final Map<ColIndexes, CompressedSizeInfoColGroup> mem;\n+ private int st1 = 0, st2 = 0, st3 = 0, st4 = 0;\n+\n+ public Memorizer() {\n+ mem = new HashMap<>();\n+ }\n+\n+ public void put(CompressedSizeInfoColGroup g) {\n+ mem.put(new ColIndexes(g.getColumns()), g);\n+ }\n+\n+ public CompressedSizeInfoColGroup get(CompressedSizeInfoColGroup g) {\n+ return mem.get(new ColIndexes(g.getColumns()));\n+ }\n+\n+ public CompressedSizeInfoColGroup get(int[] c) {\n+ return mem.get(new ColIndexes(c));\n+ }\n+\n+ public CompressedSizeInfoColGroup getOrCreate(int[] c1, int[] c2) {\n+ final int[] c = Util.join(c1, c2);\n+ final ColIndexes cI = new ColIndexes(Util.join(c1, c2));\n+ CompressedSizeInfoColGroup g = mem.get(cI);\n+ st2++;\n+ if(g == null) {\n+ final CompressedSizeInfoColGroup left = mem.get(new ColIndexes(c1));\n+ final CompressedSizeInfoColGroup right = mem.get(new ColIndexes(c2));\n+ final boolean leftConst = left.getBestCompressionType(_cs) == CompressionType.CONST &&\n+ left.getNumOffs() == 0;\n+ final boolean rightConst = right.getBestCompressionType(_cs) == CompressionType.CONST &&\n+ right.getNumOffs() == 0;\n+ if(leftConst)\n+ g = CompressedSizeInfoColGroup.addConstGroup(c, right, _cs.validCompressions);\n+ else if(rightConst)\n+ g = CompressedSizeInfoColGroup.addConstGroup(c, left, _cs.validCompressions);\n+ else {\n+ st3++;\n+ g = _est.estimateJoinCompressedSize(left, right);\n+ }\n+\n+ if(leftConst || rightConst)\n+ st4++;\n+\n+ mem.put(cI, g);\n+ }\n+ return g;\n+ }\n+\n+ public void incst1() {\n+ st1++;\n+ }\n+\n+ public String stats() {\n+ return st1 + \" \" + st2 + \" \" + st3 + \" \" + st4;\n+ }\n+\n+ public void resetStats() {\n+ st1 = 0;\n+ st2 = 0;\n+ st3 = 0;\n+ st4 = 0;\n+ }\n+\n+ @Override\n+ public String toString() {\n+ return mem.toString();\n+ }\n+ }\n+\n+ private static class ColIndexes {\n+ final int[] _indexes;\n+\n+ public ColIndexes(int[] indexes) {\n+ _indexes = indexes;\n+ }\n+\n+ @Override\n+ public int hashCode() {\n+ return Arrays.hashCode(_indexes);\n+ }\n+\n+ @Override\n+ public boolean equals(Object that) {\n+ ColIndexes thatGrp = (ColIndexes) that;\n+ return Arrays.equals(_indexes, thatGrp._indexes);\n+ }\n+ }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/cocode/CoCodeCost.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/cocode/CoCodeCost.java", "diff": "@@ -61,17 +61,14 @@ public class CoCodeCost extends AColumnCoCoder {\n}\nprivate List<CompressedSizeInfoColGroup> join(List<CompressedSizeInfoColGroup> currentGroups) {\n- // return joinToSmallForAnalysis(currentGroups);\n- List<CompressedSizeInfoColGroup> filteredGroups = joinToSmallForAnalysis(currentGroups);\n- // return currentGroups;\nComparator<CompressedSizeInfoColGroup> comp = Comparator.comparing(CompressedSizeInfoColGroup::getNumVals);\n- Queue<CompressedSizeInfoColGroup> que = new PriorityQueue<>(filteredGroups.size(), comp);\n+ Queue<CompressedSizeInfoColGroup> que = new PriorityQueue<>(currentGroups.size(), comp);\nList<CompressedSizeInfoColGroup> ret = new ArrayList<>();\n- for(CompressedSizeInfoColGroup g : filteredGroups) {\n+ for(CompressedSizeInfoColGroup g : currentGroups)\nif(g != null)\nque.add(g);\n- }\n+\nCompressedSizeInfoColGroup l = que.poll();\n@@ -96,7 +93,7 @@ public class CoCodeCost extends AColumnCoCoder {\nl = que.poll();\n}\n-\n+ if(l != null)\nret.add(l);\nfor(CompressedSizeInfoColGroup g : que)\n@@ -104,24 +101,4 @@ public class CoCodeCost extends AColumnCoCoder {\nreturn ret;\n}\n-\n- private List<CompressedSizeInfoColGroup> joinToSmallForAnalysis(List<CompressedSizeInfoColGroup> currentGroups) {\n- return currentGroups;\n- // List<CompressedSizeInfoColGroup> tmp = new ArrayList<>();\n- // int id = 0;\n- // while(id < currentGroups.size() - 1) {\n- // CompressedSizeInfoColGroup g1 = currentGroups.get(id);\n- // CompressedSizeInfoColGroup g2 = currentGroups.get(id + 1);\n- // if(g1.getNumVals() * g2.getNumVals() < toSmallForAnalysis) {\n- // tmp.add(joinWithoutAnalysis(g1, g2));\n- // }\n- // else {\n- // tmp.add(g1);\n- // tmp.add(g2);\n- // }\n- // id += 2;\n-\n- // }\n- // return tmp;\n- }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/cocode/CoCodeCostMatrixMult.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/cocode/CoCodeCostMatrixMult.java", "diff": "@@ -64,32 +64,27 @@ public class CoCodeCostMatrixMult extends AColumnCoCoder {\nList<CompressedSizeInfoColGroup> ret = new ArrayList<>();\nfor(CompressedSizeInfoColGroup g : currentGroups)\n+ if(g != null)\nque.add(new CostOfJoin(g));\n- while(true) {\n- if(que.peek() != null) {\n- final CostOfJoin l = que.poll();\n- if(que.peek() != null) {\n- final CostOfJoin r = que.poll();\n+ CostOfJoin l = que.poll();\n+\n+ while(que.peek() != null) {\n+ final CostOfJoin r = que.peek();\nfinal double costIndividual = (l.cost + r.cost);\nfinal CostOfJoin g = new CostOfJoin(joinWithAnalysis(l.elm, r.elm));\n+ if(g.cost < costIndividual) {\nif(LOG.isDebugEnabled())\nLOG.debug(\"\\nl: \" + l + \"\\nr: \" + r + \"\\njoined: \" + g);\n- if(g.cost < costIndividual)\n+ que.poll();\nque.add(g);\n- else {\n- ret.add(l.elm);\n- que.add(r);\n- }\n- }\n- else {\n- ret.add(l.elm);\n- break;\n- }\n}\nelse\n- break;\n+ ret.add(l.elm);\n+ l = que.poll();\n}\n+ if(l != null)\n+ ret.add(l.elm);\nfor(CostOfJoin g : que)\nret.add(g.elm);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/cocode/CoCodeCostTSMM.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/cocode/CoCodeCostTSMM.java", "diff": "@@ -60,20 +60,8 @@ public class CoCodeCostTSMM extends AColumnCoCoder {\nprivate List<CompressedSizeInfoColGroup> join(List<CompressedSizeInfoColGroup> currentGroups) {\n- Queue<CompressedSizeInfoColGroup> que = new PriorityQueue<>(currentGroups.size(),\n- new Comparator<CompressedSizeInfoColGroup>() {\n- @Override\n- public int compare(CompressedSizeInfoColGroup a, CompressedSizeInfoColGroup b) {\n- final int aNV = a.getNumVals();\n- final int bNV = b.getNumVals();\n- if(aNV == bNV)\n- return 0;\n- else if(aNV > bNV)\n- return 1;\n- else\n- return -1;\n- }\n- });\n+ Comparator<CompressedSizeInfoColGroup> comp = Comparator.comparing(CompressedSizeInfoColGroup::getNumVals);\n+ Queue<CompressedSizeInfoColGroup> que = new PriorityQueue<>(currentGroups.size(), comp);\nList<CompressedSizeInfoColGroup> ret = new ArrayList<>();\nfor(CompressedSizeInfoColGroup g : currentGroups)\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/cocode/PlanningCoCoder.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/cocode/PlanningCoCoder.java", "diff": "package org.apache.sysds.runtime.compress.cocode;\n-import java.util.ArrayList;\n-import java.util.Arrays;\n-import java.util.HashMap;\n-import java.util.List;\n-import java.util.Map;\n-\nimport org.apache.commons.logging.Log;\nimport org.apache.commons.logging.LogFactory;\nimport org.apache.sysds.runtime.compress.CompressionSettings;\n-import org.apache.sysds.runtime.compress.colgroup.AColGroup.CompressionType;\nimport org.apache.sysds.runtime.compress.estim.CompressedSizeEstimator;\nimport org.apache.sysds.runtime.compress.estim.CompressedSizeInfo;\n-import org.apache.sysds.runtime.compress.estim.CompressedSizeInfoColGroup;\n-import org.apache.sysds.runtime.compress.utils.Util;\npublic class PlanningCoCoder {\n@@ -70,33 +61,10 @@ public class PlanningCoCoder {\npublic static CompressedSizeInfo findCoCodesByPartitioning(CompressedSizeEstimator est, CompressedSizeInfo colInfos,\nint numRows, int k, CompressionSettings cs) {\n- // establish memo table for extracted column groups\n- Memorizer mem = null;\n- List<CompressedSizeInfoColGroup> constantGroups = null;\n- if(cs.columnPartitioner == PartitionerType.BIN_PACKING) {\n- constantGroups = new ArrayList<>();\n- List<CompressedSizeInfoColGroup> newGroups = new ArrayList<>();\n- mem = new Memorizer();\n- for(CompressedSizeInfoColGroup g : colInfos.getInfo()) {\n- if(g.getBestCompressionType(cs) == CompressionType.CONST)\n- constantGroups.add(g);\n- else {\n- mem.put(g);\n- newGroups.add(g);\n- }\n- }\n- colInfos.setInfo(newGroups);\n- }\n-\n// Use column group partitioner to create partitions of columns\nCompressedSizeInfo bins = createColumnGroupPartitioner(cs.columnPartitioner, est, cs, numRows)\n.coCodeColumns(colInfos, k);\n- if(cs.columnPartitioner == PartitionerType.BIN_PACKING) {\n- getCoCodingGroupsBruteForce(bins, k, est, mem, cs);\n- bins.getInfo().addAll(constantGroups);\n- }\n-\nreturn bins;\n}\n@@ -117,189 +85,4 @@ public class PlanningCoCoder {\nthrow new RuntimeException(\"Unsupported column group partitioner: \" + type.toString());\n}\n}\n-\n- /**\n- * This methods verifies the coCoded bins actually are the best combinations within each individual Group based on\n- * the sample.\n- *\n- * @param bins The bins constructed based on lightweight estimations\n- * @param k The number of threads allowed to be used.\n- * @param est The Estimator to be used.\n- * @return\n- */\n- private static CompressedSizeInfo getCoCodingGroupsBruteForce(CompressedSizeInfo bins, int k,\n- CompressedSizeEstimator est, Memorizer mem, CompressionSettings cs) {\n-\n- List<CompressedSizeInfoColGroup> finalGroups = new ArrayList<>();\n- // For each bin of columns that is allowed to potentially cocode.\n- for(CompressedSizeInfoColGroup bin : bins.getInfo()) {\n- final int len = bin.getColumns().length;\n- if(len == 0)\n- continue;\n- else if(len == 1)\n- // early termination\n- finalGroups.add(bin);\n- else\n- finalGroups.addAll(coCodeBruteForce(bin, est, mem, cs));\n- }\n-\n- bins.setInfo(finalGroups);\n- return bins;\n- }\n-\n- private static List<CompressedSizeInfoColGroup> coCodeBruteForce(CompressedSizeInfoColGroup bin,\n- CompressedSizeEstimator est, Memorizer mem, CompressionSettings cs) {\n-\n- List<int[]> workset = new ArrayList<>(bin.getColumns().length);\n-\n- for(int i = 0; i < bin.getColumns().length; i++)\n- workset.add(new int[] {bin.getColumns()[i]});\n-\n- // process merging iterations until no more change\n- while(workset.size() > 1) {\n- long changeInSize = 0;\n- CompressedSizeInfoColGroup tmp = null;\n- int[] selected1 = null, selected2 = null;\n- for(int i = 0; i < workset.size(); i++) {\n- for(int j = i + 1; j < workset.size(); j++) {\n- final int[] c1 = workset.get(i);\n- final int[] c2 = workset.get(j);\n- final long sizeC1 = mem.get(c1).getMinSize();\n- final long sizeC2 = mem.get(c2).getMinSize();\n-\n- mem.incst1();\n- // pruning filter : skip dominated candidates\n- // Since even if the entire size of one of the column lists is removed,\n- // it still does not improve compression\n- if(-Math.min(sizeC1, sizeC2) > changeInSize)\n- continue;\n-\n- // Join the two column groups.\n- // and Memorize the new join.\n- final CompressedSizeInfoColGroup c1c2Inf = mem.getOrCreate(c1, c2, est, cs);\n- final long sizeC1C2 = c1c2Inf.getMinSize();\n-\n- long newSizeChangeIfSelected = sizeC1C2 - sizeC1 - sizeC2;\n- // Select the best join of either the currently selected\n- // or keep the old one.\n- if((tmp == null && newSizeChangeIfSelected < changeInSize) || tmp != null &&\n- (newSizeChangeIfSelected < changeInSize || newSizeChangeIfSelected == changeInSize &&\n- c1c2Inf.getColumns().length < tmp.getColumns().length)) {\n- changeInSize = newSizeChangeIfSelected;\n- tmp = c1c2Inf;\n- selected1 = c1;\n- selected2 = c2;\n- }\n- }\n- }\n-\n- if(tmp != null) {\n- workset.remove(selected1);\n- workset.remove(selected2);\n- workset.add(tmp.getColumns());\n- }\n- else\n- break;\n- }\n-\n- LOG.debug(mem.stats());\n- mem.resetStats();\n-\n- List<CompressedSizeInfoColGroup> ret = new ArrayList<>(workset.size());\n-\n- for(int[] w : workset)\n- ret.add(mem.get(w));\n-\n- return ret;\n- }\n-\n- public static class Memorizer {\n- private final Map<ColIndexes, CompressedSizeInfoColGroup> mem;\n- private int st1 = 0, st2 = 0, st3 = 0, st4 = 0;\n-\n- public Memorizer() {\n- mem = new HashMap<>();\n- }\n-\n- public void put(CompressedSizeInfoColGroup g) {\n- mem.put(new ColIndexes(g.getColumns()), g);\n- }\n-\n- public CompressedSizeInfoColGroup get(CompressedSizeInfoColGroup g) {\n- return mem.get(new ColIndexes(g.getColumns()));\n- }\n-\n- public CompressedSizeInfoColGroup get(int[] c) {\n- return mem.get(new ColIndexes(c));\n- }\n-\n- public CompressedSizeInfoColGroup getOrCreate(int[] c1, int[] c2, CompressedSizeEstimator est,\n- CompressionSettings cs) {\n- final int[] c = Util.join(c1, c2);\n- final ColIndexes cI = new ColIndexes(Util.join(c1, c2));\n- CompressedSizeInfoColGroup g = mem.get(cI);\n- st2++;\n- if(g == null) {\n- final CompressedSizeInfoColGroup left = mem.get(new ColIndexes(c1));\n- final CompressedSizeInfoColGroup right = mem.get(new ColIndexes(c2));\n- final boolean leftConst = left.getBestCompressionType(cs) == CompressionType.CONST &&\n- left.getNumOffs() == 0;\n- final boolean rightConst = right.getBestCompressionType(cs) == CompressionType.CONST &&\n- right.getNumOffs() == 0;\n- if(leftConst)\n- g = CompressedSizeInfoColGroup.addConstGroup(c, right, cs.validCompressions);\n- else if(rightConst)\n- g = CompressedSizeInfoColGroup.addConstGroup(c, left, cs.validCompressions);\n- else {\n- st3++;\n- g = est.estimateJoinCompressedSize(left, right);\n- }\n-\n- if(leftConst || rightConst)\n- st4++;\n-\n- mem.put(cI, g);\n- }\n- return g;\n- }\n-\n- public void incst1() {\n- st1++;\n- }\n-\n- public String stats() {\n- return st1 + \" \" + st2 + \" \" + st3 + \" \" + st4;\n- }\n-\n- public void resetStats() {\n- st1 = 0;\n- st2 = 0;\n- st3 = 0;\n- st4 = 0;\n- }\n-\n- @Override\n- public String toString() {\n- return mem.toString();\n- }\n- }\n-\n- private static class ColIndexes {\n- final int[] _indexes;\n-\n- public ColIndexes(int[] indexes) {\n- _indexes = indexes;\n- }\n-\n- @Override\n- public int hashCode() {\n- return Arrays.hashCode(_indexes);\n- }\n-\n- @Override\n- public boolean equals(Object that) {\n- ColIndexes thatGrp = (ColIndexes) that;\n- return Arrays.equals(_indexes, thatGrp._indexes);\n- }\n- }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/compress/compressScale.java", "new_path": "src/test/java/org/apache/sysds/test/functions/compress/compressScale.java", "diff": "@@ -91,11 +91,11 @@ public class compressScale extends AutomatedTestBase {\ndouble outStd = Double.parseDouble(runTest(null).toString().split(\"\\n\")[0].split(\" \")[0]);\nLOG.debug(\"ULA : \" + outStd);\n- programArgs[1] = configPath(\"SystemDS-config-compress-cost-RLE.xml\");\n- double RLEoutC = Double.parseDouble(runTest(null).toString().split(\"\\n\")[0].split(\" \")[0]);\n- assertTrue(DMLCompressionStatistics.haveCompressed());\n- DMLCompressionStatistics.reset();\n- LOG.debug(\"RLE : \" + RLEoutC);\n+ // programArgs[1] = configPath(\"SystemDS-config-compress-cost-RLE.xml\");\n+ // double RLEoutC = Double.parseDouble(runTest(null).toString().split(\"\\n\")[0].split(\" \")[0]);\n+ // assertTrue(DMLCompressionStatistics.haveCompressed());\n+ // DMLCompressionStatistics.reset();\n+ // LOG.debug(\"RLE : \" + RLEoutC);\nprogramArgs[1] = configPath(\"SystemDS-config-compress-cost-OLE.xml\");\ndouble OLEOutC = Double.parseDouble(runTest(null).toString().split(\"\\n\")[0].split(\" \")[0]);\n@@ -116,7 +116,7 @@ public class compressScale extends AutomatedTestBase {\nLOG.debug(\"CLA : \" + ALLoutC);\nassertEquals(outStd, OLEOutC, 0.1);\n- assertEquals(outStd, RLEoutC, 0.1);\n+ // assertEquals(outStd, RLEoutC, 0.1);\nassertEquals(outStd, DDCoutC, 0.1);\nassertEquals(outStd, ALLoutC, 0.1);\n" } ]

[ { "change_type": "MODIFY", "old_path": "dev/release/src/test/java/org/apache/sysds/validation/ValidateLicAndNotice.java", "new_path": "dev/release/src/test/java/org/apache/sysds/validation/ValidateLicAndNotice.java", "diff": "@@ -60,7 +60,6 @@ public class ValidateLicAndNotice\nstatic final String[][] packageLicenses =\n{ {\"org/antlr\", \"ANTLR 4 Runtime (http://www.antlr.org/antlr4-runtime) org.antlr:antlr4-runtime:4.5.3\"},\n{\"org/apache/wink/json4j\",\"Apache Wink :: JSON4J (http://www.apache.org/wink/wink-json4j/) org.apache.wink:wink-json4j:1.4\"},\n- {\"caffe\",\"The proto file (src/main/proto/caffe/caffe.proto) is part of Caffe project,\"},\n{\"org/tensorflow\",\"The proto files (src/main/proto/tensorflow/event.proto and src/main/proto/tensorflow/summary.proto) is part of TensorFlow project,\"},\n{\"jcuda\",\"JCuda (jcuda.org)\"},\n};\n" }, { "change_type": "MODIFY", "old_path": "src/assembly/extra/LICENSE", "new_path": "src/assembly/extra/LICENSE", "diff": "===============================================================================\n-The proto file (src/main/proto/caffe/caffe.proto) is part of Caffe project,\n-which is used to generate caffe java package.\n-Caffe are distributed under the below license.\n-\n-COPYRIGHT\n-\n-All contributions by the University of California:\n-Copyright (c) 2014-2017 The Regents of the University of California (Regents)\n-All rights reserved.\n-\n-All other contributions:\n-Copyright (c) 2014-2017, the respective contributors\n-All rights reserved.\n-\n-Caffe uses a shared copyright model: each contributor holds copyright over\n-their contributions to Caffe. The project versioning records all such\n-contribution and copyright details. If a contributor wants to further mark\n-their specific copyright on a particular contribution, they should indicate\n-their copyright solely in the commit message of the change when it is\n-committed.\n-\n-LICENSE\n-\n-Redistribution and use in source and binary forms, with or without\n-modification, are permitted provided that the following conditions are met:\n-\n-1. Redistributions of source code must retain the above copyright notice, this\n- list of conditions and the following disclaimer.\n-2. Redistributions in binary form must reproduce the above copyright notice,\n- this list of conditions and the following disclaimer in the documentation\n- and/or other materials provided with the distribution.\n-\n-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS \"AS IS\" AND\n-ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED\n-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE\n-DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR\n-ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES\n-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;\n-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND\n-ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT\n-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS\n-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.\n-\n-CONTRIBUTION AGREEMENT\n-\n-By contributing to the BVLC/caffe repository through pull-request, comment,\n-or otherwise, the contributor releases their content to the\n-license and copyright terms herein.\n-\n-===============================================================================\n-\nThe following compile-scope dependencies come under the MIT License\nJCuda (jcuda.org)\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/codegen/SpoofCompiler.java", "new_path": "src/main/java/org/apache/sysds/hops/codegen/SpoofCompiler.java", "diff": "@@ -728,12 +728,15 @@ public class SpoofCompiler {\n//generate cplan for existing memo table entry\nif( memo.containsTopLevel(hop.getHopID()) ) {\n- cplans.put(hop.getHopID(), TemplateUtils\n+ Pair<Hop[],CNodeTpl> tmp = TemplateUtils\n.createTemplate(memo.getBest(hop.getHopID()).type)\n- .constructCplan(hop, memo, compileLiterals));\n+ .constructCplan(hop, memo, compileLiterals);\n+ if( tmp != null ) {\n+ cplans.put(hop.getHopID(), tmp);\nif (DMLScript.STATISTICS)\nStatistics.incrementCodegenCPlanCompile(1);\n}\n+ }\n//process children recursively, but skip compiled operator\nif( cplans.containsKey(hop.getHopID()) ) {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/codegen/template/TemplateCell.java", "new_path": "src/main/java/org/apache/sysds/hops/codegen/template/TemplateCell.java", "diff": "@@ -149,10 +149,16 @@ public class TemplateCell extends TemplateBase\n.filter(h -> !(h.getDataType().isScalar() && tmp.get(h.getHopID()).isLiteral()))\n.sorted(new HopInputComparator()).toArray(Hop[]::new);\n- //construct template node\n+ //prepare input nodes\nArrayList<CNode> inputs = new ArrayList<>();\nfor( Hop in : sinHops )\ninputs.add(tmp.get(in.getHopID()));\n+\n+ //sanity check for pure scalar inputs\n+ if( inputs.stream().allMatch(h -> h.getDataType().isScalar()) )\n+ return null; //later eliminated by cleanupCPlans\n+\n+ //construct template node\nCNode output = tmp.get(hop.getHopID());\nCNodeCell tpl = new CNodeCell(inputs, output);\ntpl.setCellType(TemplateUtils.getCellType(hop));\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/builtin/BuiltinGridSearchTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/builtin/BuiltinGridSearchTest.java", "diff": "@@ -22,11 +22,12 @@ package org.apache.sysds.test.functions.builtin;\nimport org.junit.Assert;\nimport org.junit.Test;\n+import java.io.File;\n+\nimport org.apache.sysds.common.Types.ExecMode;\nimport org.apache.sysds.test.AutomatedTestBase;\nimport org.apache.sysds.test.TestConfiguration;\nimport org.apache.sysds.test.TestUtils;\n-import org.apache.sysds.utils.Statistics;\npublic class BuiltinGridSearchTest extends AutomatedTestBase\n{\n@@ -39,6 +40,7 @@ public class BuiltinGridSearchTest extends AutomatedTestBase\nprivate final static int rows = 400;\nprivate final static int cols = 20;\n+ private boolean _codegen = false;\n@Override\npublic void setUp() {\n@@ -50,52 +52,64 @@ public class BuiltinGridSearchTest extends AutomatedTestBase\n@Test\npublic void testGridSearchLmCP() {\n- runGridSearch(TEST_NAME1, ExecMode.SINGLE_NODE);\n+ runGridSearch(TEST_NAME1, ExecMode.SINGLE_NODE, false);\n}\n@Test\npublic void testGridSearchLmHybrid() {\n- runGridSearch(TEST_NAME1, ExecMode.HYBRID);\n+ runGridSearch(TEST_NAME1, ExecMode.HYBRID, false);\n+ }\n+\n+ @Test\n+ public void testGridSearchLmCodegenCP() {\n+ runGridSearch(TEST_NAME1, ExecMode.SINGLE_NODE, true);\n+ }\n+\n+ @Test\n+ public void testGridSearchLmCodegenHybrid() {\n+ runGridSearch(TEST_NAME1, ExecMode.HYBRID, true);\n}\n@Test\npublic void testGridSearchLmSpark() {\n- runGridSearch(TEST_NAME1, ExecMode.SPARK);\n+ runGridSearch(TEST_NAME1, ExecMode.SPARK, false);\n}\n@Test\npublic void testGridSearchMLogregCP() {\n- runGridSearch(TEST_NAME2, ExecMode.SINGLE_NODE);\n+ runGridSearch(TEST_NAME2, ExecMode.SINGLE_NODE, false);\n}\n@Test\npublic void testGridSearchMLogregHybrid() {\n- runGridSearch(TEST_NAME2, ExecMode.HYBRID);\n+ runGridSearch(TEST_NAME2, ExecMode.HYBRID, false);\n}\n@Test\npublic void testGridSearchLm2CP() {\n- runGridSearch(TEST_NAME3, ExecMode.SINGLE_NODE);\n+ runGridSearch(TEST_NAME3, ExecMode.SINGLE_NODE, false);\n}\n@Test\npublic void testGridSearchLm2Hybrid() {\n- runGridSearch(TEST_NAME3, ExecMode.HYBRID);\n+ runGridSearch(TEST_NAME3, ExecMode.HYBRID, false);\n}\n@Test\npublic void testGridSearchLmCvCP() {\n- runGridSearch(TEST_NAME4, ExecMode.SINGLE_NODE);\n+ runGridSearch(TEST_NAME4, ExecMode.SINGLE_NODE, false);\n}\n@Test\npublic void testGridSearchLmCvHybrid() {\n- runGridSearch(TEST_NAME4, ExecMode.HYBRID);\n+ runGridSearch(TEST_NAME4, ExecMode.HYBRID, false);\n}\n- private void runGridSearch(String testname, ExecMode et)\n+ private void runGridSearch(String testname, ExecMode et, boolean codegen)\n{\nExecMode modeOld = setExecMode(et);\n+ _codegen = codegen;\n+\ntry {\nloadTestConfiguration(getTestConfiguration(testname));\nString HOME = SCRIPT_DIR + TEST_DIR;\n@@ -111,11 +125,21 @@ public class BuiltinGridSearchTest extends AutomatedTestBase\n//expected loss smaller than default invocation\nAssert.assertTrue(TestUtils.readDMLBoolean(output(\"R\")));\n- if( et != ExecMode.SPARK )\n- Assert.assertEquals(0, Statistics.getNoOfExecutedSPInst());\n+ //Assert.assertEquals(0, Statistics.getNoOfExecutedSPInst());\n+ //TODO analyze influence of multiple subsequent tests\n}\nfinally {\nresetExecMode(modeOld);\n}\n}\n+\n+ /**\n+ * Override default configuration with custom test configuration to ensure\n+ * scratch space and local temporary directory locations are also updated.\n+ */\n+ @Override\n+ protected File getConfigTemplateFile() {\n+ return !_codegen ? super.getConfigTemplateFile() :\n+ getCodegenConfigFile(SCRIPT_DIR + \"functions/codegenalg/\", CodegenTestType.DEFAULT);\n+ }\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/systemds/examples/tutorials/mnist.py", "new_path": "src/main/python/systemds/examples/tutorials/mnist.py", "diff": "@@ -29,6 +29,9 @@ import struct\nimport numpy as np\nimport requests\n+__all__ = [\"DataManager\"]\n+\n+\nclass DataManager:\n_train_data_url: str\n@@ -42,10 +45,10 @@ class DataManager:\n_test_labels_loc: str\ndef __init__(self):\n- self._train_data_url = \"http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz\"\n- self._train_labels_url = \"http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz\"\n- self._test_data_url = \"http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz\"\n- self._test_labels_url = \"http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz\"\n+ self._train_data_url = \"https://systemds.apache.org/assets/datasets/mnist/train-images-idx3-ubyte.gz\"\n+ self._train_labels_url = \"https://systemds.apache.org/assets/datasets/mnist/train-labels-idx1-ubyte.gz\"\n+ self._test_data_url = \"https://systemds.apache.org/assets/datasets/mnist/t10k-images-idx3-ubyte.gz\"\n+ self._test_labels_url = \"https://systemds.apache.org/assets/datasets/mnist/t10k-labels-idx1-ubyte.gz\"\nself._train_data_loc = \"systemds/examples/tutorials/mnist/train_data.gz\"\nself._train_labels_loc = \"systemds/examples/tutorials/mnist/train_labels.gz\"\n" }, { "change_type": "DELETE", "old_path": "src/main/python/systemds/examples/tutorials/sherlockData.py", "new_path": null, "diff": "-#!/usr/bin/env python3\n-# -------------------------------------------------------------\n-#\n-# Licensed to the Apache Software Foundation (ASF) under one\n-# or more contributor license agreements. See the NOTICE file\n-# distributed with this work for additional information\n-# regarding copyright ownership. The ASF licenses this file\n-# to you under the Apache License, Version 2.0 (the\n-# \"License\"); you may not use this file except in compliance\n-# with the License. You may obtain a copy of the License at\n-#\n-# http://www.apache.org/licenses/LICENSE-2.0\n-#\n-# Unless required by applicable law or agreed to in writing,\n-# software distributed under the License is distributed on an\n-# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n-# KIND, either express or implied. See the License for the\n-# specific language governing permissions and limitations\n-# under the License.\n-#\n-# -------------------------------------------------------------\n-\n-import os\n-from google_drive_downloader import GoogleDriveDownloader as gd\n-import pandas as pd\n-\n-class SherlockData:\n- '''\n- This data set holds data for semantic data type detection.\n-\n- The data can be used to train and test the sherlock network implemented in sherlock.dml\n- The X_*.csv files have to be processed with the function: sherlock::transform_values()\n- The y_*.csv files have to be processed with the function: \\n\n- sherlock::transform_encode_labels() to encode the output categories to numbers.\\n\n- sherlock::transform_apply_labels() is used to apply the above created encoding to the remaining input files.\n- The processed files can then be used to train the network: \\n\n- sherlock(X_train, y_train)\n- '''\n-\n- _base_path: str\n- _processed_dir: str\n- _raw_dir: str\n-\n- def __init__(self):\n- self._base_path = \"systemds/examples/tutorials/sherlock/\"\n- self._processed_dir = \"data/processed/\"\n- self._raw_dir = \"data/raw/\"\n-\n- def parquet_to_csv(self, parquetfile):\n- print(f'got file: {parquetfile}')\n- df = pd.read_parquet(parquetfile)\n- dest_name = str(parquetfile).replace(\"parquet\", \"csv\")\n- df.to_csv(dest_name)\n-\n- def get_train_values(self, processed):\n- return self._get_values(processed=processed, type=\"train\")\n-\n- def get_val_values(self, processed):\n- return self._get_values(processed=processed, type=\"val\")\n-\n- def get_test_values(self, processed):\n- return self._get_values(processed=processed, type=\"test\")\n-\n- def get_train_labels(self, processed):\n- return self._get_labels(processed=processed, type=\"train\")\n-\n- def get_val_labels(self, processed):\n- return self._get_labels(processed=processed, type=\"val\")\n-\n- def get_test_labels(self, processed):\n- return self._get_labels(processed=processed, type=\"test\")\n-\n- def _get_values(self, processed, type):\n- filename_parquet = self._base_path\n- filename_parquet += self._processed_dir + \"X_{}.parquet\".format(type) \\\n- if processed ==True else self._raw_dir + \"/{}_values.parquet\".format(type)\n-\n- if not os.path.exists(filename_parquet):\n- self._download_data(self._base_path)\n- return pd.read_parquet(filename_parquet)\n-\n- def _get_labels(self, processed, type):\n- filename_parquet = self._base_path\n- filename_parquet += self._processed_dir + \"y_{}.parquet\".format(type) \\\n- if processed ==True else self._raw_dir + \"{}_labels.parquet\".format(type)\n-\n- if not os.path.exists(filename_parquet):\n- self._download_data(self._base_path)\n- return pd.read_parquet(filename_parquet)\n-\n- def _download_data(self, data_dir):\n- \"\"\"Download raw and preprocessed data files.\n- The data is downloaded from Google Drive and stored in the 'data/' directory.\n- \"\"\"\n- print(f\"Downloading the raw and preprocessed data into {data_dir}.\")\n-\n- if not os.path.exists(data_dir + \"data\"):\n- os.makedirs(data_dir, exist_ok=True)\n- print('Downloading data directory.')\n- filename = data_dir + \"data.zip\"\n- gd.download_file_from_google_drive(\n- file_id='1-g0zbKFAXz7zKZc0Dnh74uDBpZCv4YqU',\n- dest_path=filename,\n- unzip=True,\n- showsize=True\n- )\n-\n- print('Data was downloaded.')\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/transform/encode/ColumnEncoderComposite.java", "new_path": "src/main/java/org/apache/sysds/runtime/transform/encode/ColumnEncoderComposite.java", "diff": "@@ -213,9 +213,13 @@ public class ColumnEncoderComposite extends ColumnEncoder {\naddEncoder(otherEnc);\n}\n}\n- else {\n+ else\naddEncoder(other);\n+\n+ updateAllDCEncoders();\n}\n+\n+ public void updateAllDCEncoders(){\n// update dummycode encoder domain sizes based on distinctness information from other encoders\nColumnEncoderDummycode dc = getEncoder(ColumnEncoderDummycode.class);\nif(dc != null)\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/transform/encode/MultiColumnEncoder.java", "new_path": "src/main/java/org/apache/sysds/runtime/transform/encode/MultiColumnEncoder.java", "diff": "@@ -83,8 +83,13 @@ public class MultiColumnEncoder implements Encoder {\nMatrixBlock out;\ntry {\nbuild(in, k);\n+ if(_legacyMVImpute != null){\n+ // These operations are redundant for every encoder excluding the legacyMVImpute, the workaround to fix\n+ // it for this encoder would be very dirty. This will only have a performance impact if there is a lot of\n+ // recoding in combination with the legacyMVImpute. But since it is legacy this should be fine\n_meta = getMetaData(new FrameBlock(in.getNumColumns(), Types.ValueType.STRING));\ninitMetaData(_meta);\n+ }\n// apply meta data\nout = apply(in, k);\n}\n@@ -104,8 +109,10 @@ public class MultiColumnEncoder implements Encoder {\nbuildMT(in, k);\n}\nelse {\n- for(ColumnEncoder columnEncoder : _columnEncoders)\n+ for(ColumnEncoderComposite columnEncoder : _columnEncoders){\ncolumnEncoder.build(in);\n+ columnEncoder.updateAllDCEncoders();\n+ }\n}\nlegacyBuild(in);\n}\n@@ -117,6 +124,8 @@ public class MultiColumnEncoder implements Encoder {\ntry {\nif(blockSize != in.getNumRows()) {\n// Partial builds and merges\n+ // Most of the time not worth it for RC with the current implementation, GC overhead is to large.\n+ // Depending on unique values and rows more testing need to be done\nList<List<Future<Object>>> partials = new ArrayList<>();\nfor(ColumnEncoderComposite encoder : _columnEncoders) {\nList<Callable<Object>> partialBuildTasks = encoder.getPartialBuildTasks(in, blockSize);\n@@ -124,7 +133,7 @@ public class MultiColumnEncoder implements Encoder {\npartials.add(null);\ncontinue;\n}\n- partials.add(pool.invokeAll(partialBuildTasks));\n+ partials.add(partialBuildTasks.stream().map(pool::submit).collect(Collectors.toList()));\n}\nfor(int e = 0; e < _columnEncoders.size(); e++) {\nList<Future<Object>> partial = partials.get(e);\n@@ -179,7 +188,7 @@ public class MultiColumnEncoder implements Encoder {\nif(in.getNumColumns() != numEncoders)\nthrow new DMLRuntimeException(\"Not every column in has a CompositeEncoder. Please make sure every column \"\n+ \"has a encoder or slice the input accordingly\");\n- // Denseblock allocation since access is only on the DenseBlock\n+ // Block allocation for MT access\nout.allocateBlock();\nif(out.isInSparseFormat()) {\nSparseBlock block = out.getSparseBlock();\n@@ -204,7 +213,7 @@ public class MultiColumnEncoder implements Encoder {\noffset += columnEncoder.getEncoder(ColumnEncoderDummycode.class)._domainSize - 1;\n}\n}\n- // Recomputing NNZ since we access the Dense block directly\n+ // Recomputing NNZ since we access the block directly\n// TODO set NNZ explicit count them in the encoders\nout.recomputeNonZeros();\nif(_legacyOmit != null)\n@@ -634,6 +643,8 @@ public class MultiColumnEncoder implements Encoder {\n@Override\npublic Integer call() throws Exception {\n_encoder.build(_input);\n+ if(_encoder instanceof ColumnEncoderComposite)\n+ ((ColumnEncoderComposite) _encoder).updateAllDCEncoders();\nreturn 1;\n}\n}\n@@ -652,6 +663,7 @@ public class MultiColumnEncoder implements Encoder {\n@Override\npublic Integer call() throws Exception {\n_encoder.mergeBuildPartial(_partials, 0, _partials.size());\n+ _encoder.updateAllDCEncoders();\nreturn 1;\n}\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "bin/systemds", "new_path": "bin/systemds", "diff": "@@ -39,9 +39,12 @@ if [ -z \"$SYSDS_QUIET\" ]; then\nSYSDS_QUIET=0\nfi\n-# if not set by env, set to singlenode execution by default\n+# if not set by env, set to default exec modes\nif [[ -z \"$SYSDS_EXEC_MODE\" ]]; then\n- SYSDS_EXEC_MODE=singlenode\n+ case \"$SYSDS_DISTRIBUTED\" in\n+ 0) SYSDS_EXEC_MODE=singlenode ;;\n+ *) SYSDS_EXEC_MODE=hybrid ;;\n+ esac\nfi\n# an echo toggle\n@@ -76,6 +79,7 @@ else\n--master yarn \\\n--deploy-mode client\\\n--driver-memory 96g \\\n+ --conf spark.ui.showConsoleProgress=true \\\n--num-executors 4 \\\n--executor-memory 64g \\\n--executor-cores 16 \"\n" }, { "change_type": "MODIFY", "old_path": "conf/log4j.properties", "new_path": "conf/log4j.properties", "diff": "log4j.rootLogger=ERROR,console\nlog4j.logger.org.apache.sysds=ERROR\n-log4j.logger.org.apache.spark=OFF\n+log4j.logger.org.apache.spark=ERROR\nlog4j.logger.org.apache.hadoop=OFF\nlog4j.appender.console=org.apache.log4j.ConsoleAppender\n" }, { "change_type": "MODIFY", "old_path": "conf/log4j.properties.template", "new_path": "conf/log4j.properties.template", "diff": "log4j.rootLogger=ERROR,console\nlog4j.logger.org.apache.sysds=ERROR\n-log4j.logger.org.apache.spark=OFF\n+log4j.logger.org.apache.spark=ERROR\nlog4j.logger.org.apache.hadoop=OFF\nlog4j.appender.console=org.apache.log4j.ConsoleAppender\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/assembly/bin.xml", "new_path": "src/assembly/bin.xml", "diff": "<excludes>\n<exclude>algorithms/obsolete/*</exclude>\n<exclude>algorithms/obsolete</exclude>\n- <exclude>perftest/*</exclude>\n+ <exclude>perftest/**/*</exclude>\n<exclude>perftest</exclude>\n+ <exclude>perftestDeprecated/*</exclude>\n+ <exclude>perftestDeprecated</exclude>\n<exclude>staging/**/*</exclude>\n<exclude>staging</exclude>\n</excludes>\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "<artifactId>protobuf-java-util</artifactId>\n<version>3.12.2</version>\n</dependency>\n+\n+ <dependency>\n+ <groupId>org.apache.maven.plugins</groupId>\n+ <artifactId>maven-assembly-plugin</artifactId>\n+ <version>3.3.0</version>\n+ </dependency>\n+\n</dependencies>\n</project>\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/systemds/operator/nodes/source.py", "new_path": "src/main/python/systemds/operator/nodes/source.py", "diff": "@@ -191,9 +191,8 @@ class Source(OperationNode):\nreturn filtered_lines\n- def code_line(self, unnamed_input_vars: Sequence[str], named_input_vars: Dict[str, str]) -> str:\n+ def code_line(self, var_name: str, unnamed_input_vars: Sequence[str], named_input_vars: Dict[str, str]) -> str:\nline = f'source({self.operation}) as { self.__name}'\n- self._already_added = True\nreturn line\ndef compute(self, verbose: bool = False, lineage: bool = False):\n" }, { "change_type": "MODIFY", "old_path": "src/main/python/systemds/script_building/dag.py", "new_path": "src/main/python/systemds/script_building/dag.py", "diff": "@@ -101,7 +101,6 @@ class DAGNode(ABC):\n_output_type: OutputType\n_script: Optional[\"DMLScript\"]\n_is_python_local_data: bool\n- _already_added: bool\n_dml_name: str\ndef compute(self, verbose: bool = False, lineage: bool = False) -> Any:\n@@ -160,10 +159,6 @@ class DAGNode(ABC):\ndef output_type(self):\nreturn self._output_type\n- @property\n- def already_added(self):\n- return self._already_added\n-\n@property\ndef script(self):\nreturn self._script\n" }, { "change_type": "MODIFY", "old_path": "src/main/python/systemds/script_building/script.py", "new_path": "src/main/python/systemds/script_building/script.py", "diff": "@@ -185,11 +185,6 @@ class DMLScript:\nif dag_node.dml_name != \"\":\nreturn dag_node.dml_name\n- if dag_node._output_type == OutputType.IMPORT:\n- if not dag_node.already_added:\n- self.add_code(dag_node.code_line(None, None))\n- return None\n-\nif dag_node._source_node is not None:\nself._dfs_dag_nodes(dag_node._source_node)\n# for each node do the dfs operation and save the variable names in `input_var_names`\n@@ -228,6 +223,8 @@ class DMLScript:\nself._dfs_clear_dag_nodes(n)\nfor name, n in dag_node.named_input_nodes.items():\nself._dfs_clear_dag_nodes(n)\n+ if dag_node._source_node is not None:\n+ self._dfs_clear_dag_nodes(dag_node._source_node)\ndef _next_unique_var(self) -> str:\n\"\"\"Gets the next unique variable name\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/python/tests/source/test_source_reuse.py", "diff": "+# -------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+# -------------------------------------------------------------\n+\n+import unittest\n+\n+import numpy as np\n+from systemds.context import SystemDSContext\n+\n+\n+class TestSourceReuse(unittest.TestCase):\n+\n+ sds: SystemDSContext = None\n+ source_reuse = None\n+\n+ @classmethod\n+ def setUpClass(cls):\n+ cls.sds = SystemDSContext()\n+ cls.source_reuse = cls.sds.source(\"./tests/source/source_01.dml\",\n+ \"test\")\n+\n+ @classmethod\n+ def tearDownClass(cls):\n+ cls.sds.close()\n+\n+ def test_01_single_call(self):\n+ self.call()\n+\n+ def test_02_second_call(self):\n+ self.call()\n+\n+ def test_03_same_function(self):\n+ s = self.sds.source(\"./tests/source/source_01.dml\",\n+ \"test\")\n+ c = s.test_01().compute()\n+ d = s.test_01().compute()\n+ self.assertTrue(np.allclose(c, d))\n+\n+ def call(self):\n+ c = self.source_reuse.test_01()\n+ res = c.compute()\n+ self.assertEqual(1, self.imports(c.script_str))\n+ self.assertTrue(np.allclose(np.array([[1]]), res))\n+\n+\n+\n+ def imports(self, script:str) -> int:\n+ return script.split(\"\\n\").count('source(\"./tests/source/source_01.dml\") as test')\n+\n+if __name__ == \"__main__\":\n+ unittest.main(exit=False)\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupFactory.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupFactory.java", "diff": "@@ -34,7 +34,6 @@ import org.apache.commons.logging.LogFactory;\nimport org.apache.sysds.runtime.DMLCompressionException;\nimport org.apache.sysds.runtime.DMLRuntimeException;\nimport org.apache.sysds.runtime.compress.CompressionSettings;\n-import org.apache.sysds.runtime.compress.cocode.PlanningCoCoder.PartitionerType;\nimport org.apache.sysds.runtime.compress.colgroup.AColGroup.CompressionType;\nimport org.apache.sysds.runtime.compress.colgroup.dictionary.ADictionary;\nimport org.apache.sysds.runtime.compress.colgroup.dictionary.Dictionary;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/ParForProgramBlock.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/ParForProgramBlock.java", "diff": "@@ -585,7 +585,7 @@ public class ParForProgramBlock extends ForProgramBlock\n//OPTIMIZATION of ParFOR body (incl all child parfor PBs)\n///////\nif( _optMode != POptMode.NONE ) {\n- // OptimizationWrapper.setLogLevel(_optLogLevel); //set optimizer log level\n+ OptimizationWrapper.setLogLevel(_optLogLevel); //set optimizer log level\nOptimizationWrapper.optimize(_optMode, sb, this, ec, _monitor); //core optimize\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/opt/OptimizationWrapper.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/opt/OptimizationWrapper.java", "diff": "@@ -25,6 +25,8 @@ import java.util.Set;\nimport org.apache.commons.logging.Log;\nimport org.apache.commons.logging.LogFactory;\n+import org.apache.log4j.Level;\n+import org.apache.log4j.Logger;\nimport org.apache.sysds.api.DMLScript;\nimport org.apache.sysds.conf.ConfigurationManager;\nimport org.apache.sysds.hops.OptimizerUtils;\n@@ -110,10 +112,10 @@ public class OptimizationWrapper\nStatisticMonitor.putPFStat( pb.getID() , Stat.OPT_T, timeVal);\n}\n- // public static void setLogLevel( Level optLogLevel ) {\n- // Logger.getLogger(\"org.apache.sysds.runtime.controlprogram.parfor.opt\")\n- // .setLevel( optLogLevel );\n- // }\n+ public static void setLogLevel( Level optLogLevel ) {\n+ Logger.getLogger(\"org.apache.sysds.runtime.controlprogram.parfor.opt\")\n+ .setLevel( optLogLevel );\n+ }\n@SuppressWarnings(\"unused\")\nprivate static void optimize( POptMode otype, int ck, double cm, ParForStatementBlock sb, ParForProgramBlock pb, ExecutionContext ec, boolean monitor )\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/VariableCPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/VariableCPInstruction.java", "diff": "@@ -1146,6 +1146,8 @@ public class VariableCPInstruction extends CPInstruction implements LineageTrace\nFileFormat fmt = ((MetaDataFormat) mo.getMetaData()).getFileFormat();\nDataCharacteristics dc = (mo.getMetaData()).getDataCharacteristics();\nif(fmt == FileFormat.HDF5 && !getInput1().getName().startsWith(org.apache.sysds.lops.Data.PREAD_PREFIX)) {\n+ //FIXME why is this writer never used?\n+ @SuppressWarnings(\"unused\")\nWriterHDF5 writer = new WriterHDF5((FileFormatPropertiesHDF5) _formatProperties);\n}\nelse {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/io/hdf5/H5BTree.java", "new_path": "src/main/java/org/apache/sysds/runtime/io/hdf5/H5BTree.java", "diff": "@@ -30,6 +30,7 @@ public class H5BTree {\nprivate static final byte[] BTREE_NODE_SIGNATURE = \"TREE\".getBytes(StandardCharsets.US_ASCII);\nprivate static final int HEADER_BYTES = 6;\n+ @SuppressWarnings(\"unused\")\nprivate final long address;\nprotected final int entriesUsed;\nprivate final long leftSiblingAddress;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/io/hdf5/H5ContiguousDataset.java", "new_path": "src/main/java/org/apache/sysds/runtime/io/hdf5/H5ContiguousDataset.java", "diff": "@@ -33,6 +33,7 @@ public class H5ContiguousDataset {\nprivate final H5RootObject rootObject;\nprivate final H5DataLayoutMessage dataLayoutMessage;\nprivate final H5DataTypeMessage dataTypeMessage;\n+ @SuppressWarnings(\"unused\")\nprivate final H5DataSpaceMessage dataSpaceMessage;\npublic H5ContiguousDataset(H5RootObject rootObject, H5ObjectHeader objectHeader) {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/io/hdf5/H5RuntimeException.java", "new_path": "src/main/java/org/apache/sysds/runtime/io/hdf5/H5RuntimeException.java", "diff": "@@ -22,6 +22,8 @@ package org.apache.sysds.runtime.io.hdf5;\nimport org.apache.sysds.runtime.DMLRuntimeException;\npublic class H5RuntimeException extends DMLRuntimeException {\n+ private static final long serialVersionUID = -3551978964353888835L;\n+\npublic H5RuntimeException(String string) {\nsuper(string);\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/component/compress/mapping/StandAloneTests.java", "new_path": "src/test/java/org/apache/sysds/test/component/compress/mapping/StandAloneTests.java", "diff": "@@ -119,7 +119,7 @@ public class StandAloneTests {\n// compare(c, new int[] {0, 1, 1, 1, 1, 0, 0, 0, 0, 0});\n}\n- private void compare(AMapToData res, int[] expected) {\n+ private static void compare(AMapToData res, int[] expected) {\nStringBuilder sb = new StringBuilder();\nsb.append(\"\\nExpected:\\n\");\nsb.append(Arrays.toString(expected));\n" } ]

[ { "change_type": "DELETE", "old_path": "scripts/builtin/getPermutations.dml", "new_path": null, "diff": "-#-------------------------------------------------------------\n-#\n-# Licensed to the Apache Software Foundation (ASF) under one\n-# or more contributor license agreements. See the NOTICE file\n-# distributed with this work for additional information\n-# regarding copyright ownership. The ASF licenses this file\n-# to you under the Apache License, Version 2.0 (the\n-# \"License\"); you may not use this file except in compliance\n-# with the License. You may obtain a copy of the License at\n-#\n-# http://www.apache.org/licenses/LICENSE-2.0\n-#\n-# Unless required by applicable law or agreed to in writing,\n-# software distributed under the License is distributed on an\n-# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n-# KIND, either express or implied. See the License for the\n-# specific language governing permissions and limitations\n-# under the License.\n-#\n-#-------------------------------------------------------------\n-\n-# Related to [SYSTEMDS-2662] dependency function for cleaning pipelines\n-# This built-in will generate all possible permutation for a given set of values\n-\n-\n-s_getPermutations = function(Frame[String] opt)\n-return(Frame[String] output)\n-{\n- idx = seq(1, ncol(opt))\n- # get the indexes of columns for recode transformation\n- index = vectorToCsv(idx)\n- # recode logical pipelines for easy handling\n- jspecR = \"{ids:true, recode:[\"+index+\"]}\";\n- [X, M] = transformencode(target=opt, spec=jspecR);\n- X = replace(target= X, pattern = NaN, replacement = 0)\n- # initialize output matrix\n- n = nrow(opt)\n- d = ncol(opt)\n- outC = matrix(0, n^d, d)\n-\n- parfor(i in 1 : d) {\n- # matrix for storing rows of ith columns\n- outR = matrix(0, 0, 1)\n- j = n^i\n- rowIdx = 1\n- for(k in 1:j) {\n- valDup = matrix(as.scalar(X[rowIdx, i]), n^(d-i), 1)\n- outR = rbind(outR, valDup)\n- rowIdx = rowIdx + 1\n- rowIdx = ifelse(((rowIdx)%%(n+1)) == 0, 1, rowIdx)\n- }\n- outC[,i] = outR\n- }\n- computeInvalid = rowMins(outC) == 0\n- outC = removeEmpty(target = outC, margin = \"rows\", select = computeInvalid == 0)\n- output = transformdecode(target=outC, spec=jspecR, meta=M);\n-}\n\\ No newline at end of file\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/util/DMVUtils.java", "new_path": "src/main/java/org/apache/sysds/runtime/util/DMVUtils.java", "diff": "@@ -49,10 +49,10 @@ public class DMVUtils {\nfor (int idx = 0; idx < numCols; idx++) {\nObject c = frame.getColumnData(idx);\n- String[] column = (String[]) c;\n+ String[] attr = (String[]) c;\nString key = \"\";\n- for (String attr : column) {\n- key = (attr.isEmpty()) ? \"NULL\": attr;\n+ for (int i = 0; i < numRows; i++) {\n+ key = (attr[i] == null) ? \"NULL\": attr[i];\naddDistinctValueOrIncrementCounter(table_Hist, key, idx);\n}\n}\n@@ -278,36 +278,38 @@ public class DMVUtils {\n{\nint row_idx = -1;\nString pattern = \"\";\n- String[] column = (String[]) col;\n- for (String attr : column) {\n+ String[] attr = (String[]) col;\n+ int numRows = frameBlock.getNumRows();\n+ for (int i = 0; i < numRows; i++) {\n+ String value = (attr[i] == null)? \"NULL\": attr[i];\nswitch (level){\ncase LEVEL1:\n- pattern = encodeRawString(attr);\n+ pattern = encodeRawString(value);\nbreak;\ncase LEVEL2:\n- pattern = encodeRawString(attr);\n+ pattern = encodeRawString(value);\npattern = removeNumbers(pattern);\nbreak;\ncase LEVEL3:\n- pattern = encodeRawString(attr);\n+ pattern = encodeRawString(value);\npattern = removeNumbers(pattern);\npattern = removeUpperLowerCase(pattern);\nbreak;\ncase LEVEL4:\n- pattern = encodeRawString(attr);\n+ pattern = encodeRawString(value);\npattern = removeNumbers(pattern);\npattern = removeUpperLowerCase(pattern);\npattern = removeInnerCharacterInPattern(pattern, DIGIT, DOT);\nbreak;\ncase LEVEL5:\n- pattern = encodeRawString(attr);\n+ pattern = encodeRawString(value);\npattern = removeNumbers(pattern);\npattern = removeUpperLowerCase(pattern);\npattern = removeInnerCharacterInPattern(pattern, DIGIT, DOT);\npattern = removeInnerCharacterInPattern(pattern, ALPHA, SPACE);\nbreak;\ncase LEVEL6:\n- pattern = encodeRawString(attr);\n+ pattern = encodeRawString(value);\npattern = removeNumbers(pattern);\npattern = removeUpperLowerCase(pattern);\npattern = removeInnerCharacterInPattern(pattern, DIGIT, DOT);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/common/Builtins.java", "new_path": "src/main/java/org/apache/sysds/common/Builtins.java", "diff": "@@ -124,7 +124,6 @@ public enum Builtins {\nFRAME_SORT(\"frameSort\", true),\nGAUSSIAN_CLASSIFIER(\"gaussianClassifier\", true),\nGET_ACCURACY(\"getAccuracy\", true),\n- GET_PERMUTATIONS(\"getPermutations\", true),\nGLM(\"glm\", true),\nGMM(\"gmm\", true),\nGMM_PREDICT(\"gmmPredict\", true),\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/assembly/bin.xml", "new_path": "src/assembly/bin.xml", "diff": "<include>*:hadoop-yarn*</include>\n<include>*:jackson-core-asl*</include>\n<include>*:jackson-mapper-asl*</include>\n+ <include>*:janino*</include>\n<include>*:log4j*</include>\n+ <include>*:netty*</include>\n<include>*:protobuf-java*</include>\n+ <include>*:py4j*</include>\n<include>*:slf4j-api*</include>\n<include>*:slf4j-log4j*</include>\n- <include>*:netty*</include>\n- <include>*:py4j*</include>\n+ <include>*:spark-core*</include>\n</includes>\n<outputDirectory>./lib</outputDirectory>\n<scope>compile</scope>\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/cuda/spoof-launcher/SpoofCUDAContext.cpp", "new_path": "src/main/cuda/spoof-launcher/SpoofCUDAContext.cpp", "diff": "#include <filesystem>\n#include <iostream>\n#include <cstdlib>\n-#include <sstream>\n-using clk = std::chrono::high_resolution_clock;\n-using sec = std::chrono::duration<double, std::ratio<1>>;\n+//#include <sstream>\n+//using clk = std::chrono::high_resolution_clock;\n+//using sec = std::chrono::duration<double, std::ratio<1>>;\nsize_t SpoofCUDAContext::initialize_cuda(uint32_t device_id, const char* resource_path) {\n@@ -56,17 +56,14 @@ size_t SpoofCUDAContext::initialize_cuda(uint32_t device_id, const char* resourc\nCUfunction func;\n- // SUM\n+ // SUM and SUM_SQ have the same behavior for intermediate buffers (squaring is done in the initial reduction step,\n+ // after that it is just summing up the temporary data)\nCHECK_CUDA(cuModuleGetFunction(&func, ctx->reductions, \"reduce_sum_f\"));\nctx->reduction_kernels_f.insert(std::make_pair(std::make_pair(SpoofOperator::AggType::FULL_AGG, SpoofOperator::AggOp::SUM), func));\n+ ctx->reduction_kernels_f.insert(std::make_pair(std::make_pair(SpoofOperator::AggType::FULL_AGG, SpoofOperator::AggOp::SUM_SQ), func));\nCHECK_CUDA(cuModuleGetFunction(&func, ctx->reductions, \"reduce_sum_d\"));\nctx->reduction_kernels_d.insert(std::make_pair(std::make_pair(SpoofOperator::AggType::FULL_AGG, SpoofOperator::AggOp::SUM), func));\n-\n- // // SUM_SQ\n- // CHECK_CUDA(cuModuleGetFunction(&func, ctx->reductions, \"reduce_sum_sq_d\"));\n- // ctx->reduction_kernels.insert(std::make_pair(\"reduce_sum_sq_d\", func));\n- // CHECK_CUDA(cuModuleGetFunction(&func, ctx->reductions, \"reduce_sum_sq_f\"));\n- // ctx->reduction_kernels.insert(std::make_pair(\"reduce_sum_sq_f\", func));\n+ ctx->reduction_kernels_d.insert(std::make_pair(std::make_pair(SpoofOperator::AggType::FULL_AGG, SpoofOperator::AggOp::SUM_SQ), func));\n// MIN\nCHECK_CUDA(cuModuleGetFunction(&func, ctx->reductions, \"reduce_min_f\"));\n@@ -83,13 +80,13 @@ size_t SpoofCUDAContext::initialize_cuda(uint32_t device_id, const char* resourc\nreturn reinterpret_cast<size_t>(ctx);\n}\n-void SpoofCUDAContext::destroy_cuda(SpoofCUDAContext *ctx, uint32_t device_id) {\n+void SpoofCUDAContext::destroy_cuda(SpoofCUDAContext *ctx, [[maybe_unused]] uint32_t device_id) {\ndelete ctx;\n// cuda device is handled by jCuda atm\n//cudaDeviceReset();\n}\n-int SpoofCUDAContext::compile(std::unique_ptr<SpoofOperator> op, const std::string &src) {\n+size_t SpoofCUDAContext::compile(std::unique_ptr<SpoofOperator> op, const std::string &src) {\n#ifndef NDEBUG\n// std::cout << \"---=== START source listing of spoof cuda kernel [ \" << name << \" ]: \" << std::endl;\n// uint32_t line_num = 0;\n" }, { "change_type": "MODIFY", "old_path": "src/main/cuda/spoof-launcher/SpoofCUDAContext.h", "new_path": "src/main/cuda/spoof-launcher/SpoofCUDAContext.h", "diff": "@@ -62,7 +62,7 @@ public:\nstatic void destroy_cuda(SpoofCUDAContext *ctx, uint32_t device_id);\n- int compile(std::unique_ptr<SpoofOperator> op, const std::string &src);\n+ size_t compile(std::unique_ptr<SpoofOperator> op, const std::string &src);\ntemplate <typename T, typename CALL>\nint launch(uint32_t opID, std::vector<Matrix<T>>& input, std::vector<Matrix<T>>& sides, Matrix<T>& output,\n" }, { "change_type": "MODIFY", "old_path": "src/main/cuda/spoof-launcher/jni_bridge.cpp", "new_path": "src/main/cuda/spoof-launcher/jni_bridge.cpp", "diff": "@@ -39,12 +39,12 @@ template<typename T>\nstruct LaunchMetadata {\nconst T& opID;\nconst T& grix;\n- const T& num_inputs;\n- const T& num_sides;\n+ const size_t& num_inputs;\n+ const size_t& num_sides;\n// num entries describing one matrix (6 entries):\n// {nnz,rows,cols,row_ptr,col_idxs,data}\n- const T& entry_size;\n+ const size_t& entry_size;\nconst T& num_scalars;\nexplicit LaunchMetadata(const size_t* jvals) : opID(jvals[0]), grix(jvals[1]), num_inputs(jvals[2]),\n@@ -58,7 +58,7 @@ Java_org_apache_sysds_hops_codegen_SpoofCompiler_initialize_1cuda_1context(\nconst char *cstr_rp = jenv->GetStringUTFChars(resource_path, nullptr);\nsize_t ctx = SpoofCUDAContext::initialize_cuda(device_id, cstr_rp);\njenv->ReleaseStringUTFChars(resource_path, cstr_rp);\n- return ctx;\n+ return static_cast<jlong>(ctx);\n}\n@@ -136,12 +136,12 @@ int launch_spoof_operator(JNIEnv *jenv, [[maybe_unused]] jclass jobj, jlong _ctx\n// wrap/cast inputs\nstd::vector<Matrix<T>> mats_in;\n- for(auto i = 0; i < meta.num_inputs; i+=meta.entry_size)\n+ for(auto i = 0ul; i < meta.num_inputs; i+=meta.entry_size)\nmats_in.emplace_back(&inputs[i]);\n// wrap/cast sides\nstd::vector<Matrix<T>> mats_sides;\n- for(auto i = 0; i < meta.num_sides; i+=meta.entry_size)\n+ for(auto i = 0ul; i < meta.num_sides; i+=meta.entry_size)\nmats_sides.emplace_back(&sides[i]);\n// wrap/cast output\n" }, { "change_type": "MODIFY", "old_path": "src/main/cuda/spoof/cellwise.cu", "new_path": "src/main/cuda/spoof/cellwise.cu", "diff": "@@ -56,9 +56,7 @@ struct SpoofCellwiseOp {\n//%NEED_RIX%\n//%NEED_CIX%\n//%NEED_GRIX%\n-\n%BODY_dense%\n-//printf(\"tid=%d a=%4.1f\\n\", threadIdx.x, a);\nreturn %OUT%;\n}\n};\n@@ -75,9 +73,4 @@ __global__ void /*%TMP%*/SPOOF_OP_NAME_SPARSE (Matrix<T>* a, Matrix<T>* b, Matri\n%AGG_OP%<T> agg_op;\nSpoofCellwiseOp<T, NUM_B> spoof_op(a, b, c, scalars, grix);\n%TYPE%_SPARSE<T, %AGG_OP%<T>, SpoofCellwiseOp<T, NUM_B>>(&(spoof_op.A), &(spoof_op.c), n, %INITIAL_VALUE%, agg_op, spoof_op);\n-\n-// if(blockIdx.x == 0 && threadIdx.x == 0) {\n-// for(auto i = 0; i < 30; ++i)\n-// printf(\"%4.3f \", spoof_op.c.val(i));\n-// }\n};\n\\ No newline at end of file\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/AggUnaryOp.java", "new_path": "src/main/java/org/apache/sysds/hops/AggUnaryOp.java", "diff": "@@ -98,7 +98,7 @@ public class AggUnaryOp extends MultiThreadedHop\nreturn false;\n}\nelse if ((_op == AggOp.SUM && (_direction == Direction.RowCol || _direction == Direction.Row || _direction == Direction.Col))\n-// || (_op == AggOp.SUM_SQ && (_direction == Direction.RowCol || _direction == Direction.Row || _direction == Direction.Col))\n+ || (_op == AggOp.SUM_SQ && (_direction == Direction.RowCol || _direction == Direction.Row || _direction == Direction.Col))\n|| (_op == AggOp.MAX && (_direction == Direction.RowCol || _direction == Direction.Row || _direction == Direction.Col))\n|| (_op == AggOp.MIN && (_direction == Direction.RowCol || _direction == Direction.Row || _direction == Direction.Col))\n|| (_op == AggOp.MEAN && (_direction == Direction.RowCol || _direction == Direction.Row || _direction == Direction.Col))\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/codegen/cplan/CNodeCell.java", "new_path": "src/main/java/org/apache/sysds/hops/codegen/cplan/CNodeCell.java", "diff": "@@ -275,8 +275,7 @@ public class CNodeCell extends CNodeTpl\n}\n@Override\npublic boolean isSupported(GeneratorAPI api) {\n- return (api == GeneratorAPI.CUDA || api == GeneratorAPI.JAVA) && _output.isSupported(api) &&\n- !(getSpoofAggOp() == SpoofCellwise.AggOp.SUM_SQ);\n+ return (api == GeneratorAPI.CUDA || api == GeneratorAPI.JAVA) && _output.isSupported(api);\n}\npublic int compile(GeneratorAPI api, String src) {\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/aggregate/SumSqTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/aggregate/SumSqTest.java", "diff": "@@ -206,7 +206,8 @@ public class SumSqTest extends AutomatedTestBase {\n// On CP and Spark modes, check that the rewrite actually\n// occurred for matrix cases and not for vector cases.\nif (rewrites && (platform == ExecType.SPARK || platform == ExecType.CP)) {\n- String prefix = (platform == ExecType.SPARK) ? Instruction.SP_INST_PREFIX : \"\";\n+ String prefix = (platform == ExecType.SPARK) ? Instruction.SP_INST_PREFIX :\n+ (DMLScript.USE_ACCELERATOR ? \"gpu_\": \"\");\nString opcode = prefix + op;\nboolean rewriteApplied = Statistics.getCPHeavyHitterOpCodes().contains(opcode);\nif (vector)\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/cuda/headers/Matrix.h", "new_path": "src/main/cuda/headers/Matrix.h", "diff": "@@ -76,27 +76,27 @@ public:\n__device__ uint32_t cols() { return _mat->cols; }\n__device__ uint32_t rows() { return _mat->rows; }\n- __device__ uint32_t len() { return _mat->data == nullptr ? len_sparse() : len_dense(); }\n+ __device__ uint32_t len() { return _mat->row_ptr == nullptr ? len_dense() : len_sparse(); }\n__device__ uint32_t pos(uint32_t rix) {\n- return _mat->data == nullptr ? pos_sparse(rix) : pos_dense(rix);\n+ return _mat->row_ptr == nullptr ? pos_dense(rix) : pos_sparse(rix);\n}\n__device__ T& val(uint32_t r, uint32_t c) {\n- return _mat->data == nullptr ? val_sparse_rc(r, c) : val_dense_rc(r,c);\n+ return _mat->row_ptr == nullptr ? val_dense_rc(r,c) : val_sparse_rc(r, c) ;\n}\n__device__ T& val(uint32_t i) {\n- return _mat->data == nullptr ? val_sparse_i(i) : val_dense_i(i);\n+ return _mat->row_ptr == nullptr ? val_dense_i(i) : val_sparse_i(i);\n}\n__device__ T& operator[](uint32_t i) { return val(i); }\n__device__ T* vals(uint32_t rix) {\n- return _mat->data == nullptr ? vals_sparse(rix) : vals_dense(rix);\n+ return _mat->row_ptr == nullptr ? vals_dense(rix) : vals_sparse(rix) ;\n}\n__device__ uint32_t row_len(uint32_t rix) {\n- return _mat->data == nullptr ? row_len_sparse(rix) : row_len_dense(rix);\n+ return _mat->row_ptr == nullptr ? row_len_dense(rix) : row_len_sparse(rix);\n}\n__device__ uint32_t* col_idxs(uint32_t rix) { return cols_sparse(rix); }\n" }, { "change_type": "MODIFY", "old_path": "src/main/cuda/headers/reduction.cuh", "new_path": "src/main/cuda/headers/reduction.cuh", "diff": "@@ -330,47 +330,30 @@ __device__ void NO_AGG_SPARSE(MatrixAccessor<T>* in, MatrixAccessor<T>* out, uin\n{\nconst uint32_t& rix = blockIdx.x;\nuint32_t tid = threadIdx.x;\n-// uint32_t rix = (gtid * VT) / in->cols();\n-// //uint32_t cix = (gtid % in->cols());// *static_cast<uint32_t>(VT);\n-// uint32_t cix = in->col_idxs(0)[gtid];\n- uint32_t row_start = in->pos(rix);\n- uint32_t row_len = in->row_len(rix);\n-\n+ uint32_t row_start = 0;\n+ uint32_t row_len = 0;\n+ if(in->hasData()) {\n+ row_start = in->pos(rix);\n+ row_len = in->row_len(rix);\n+ }\n+ else {\n+ row_start = rix * in->cols();\n+ row_len = in->cols();\n+ }\nwhile(tid < row_len) {\n+ uint32_t idx = row_start + tid;\nif(in->hasData()) {\nuint32_t *aix = in->col_idxs(rix);\nuint32_t cix = aix[tid];\n-// T result = spoof_op(in->val(rix, cix), rix*in->rows()+cix, rix, cix);\n- T result = spoof_op(in->val(row_start + tid), rix * in->rows() + cix, rix, cix);\n- out->set(row_start + tid, cix, result);\n-\n-// if(rix > 899 && rix < 903 && cix==0)\n-// if(rix < 10 && cix==0)\n-// printf(\"rix=%d row_start=%d tid=%d result=%4.3f\\n\", rix, row_start, tid, result);\n+ T result = spoof_op(in->val(idx), idx, rix, cix);\n+ out->set(idx, cix, result);\n}\nelse {\nuint32_t cix = tid;\n- T result = spoof_op(0, rix * in->rows() + cix, rix, cix);\n- out->set(row_start + tid, cix, result);\n+ T result = spoof_op(0, idx, rix, cix);\n+ out->set(idx, cix, result);\n}\ntid+=blockDim.x;\n-\n-\n-//#pragma unroll\n-// for (auto i = first_idx; i < last_idx; i++) {\n-//// out->vals(0)[i] = spoof_op(in->vals(0)[i], i);\n-//// out->col_idxs(0)[i] = gtid % blockDim.x;\n-// T result = spoof_op(in->vals(0)[i], i);\n-// out->vals(0)[i] = result;\n-// //out->col_idxs(0)[i] = i % in->cols();\n-// out->col_idxs(0)[i] = in->col_idxs(0)[i];\n-// //out->set(i/in->cols(), i%in->cols(), result);\n-// //out->set(rix, i%in->cols(), result);\n-// if (i > in->nnz() - 10)\n-// printf(\"i=%d in=%4.3f res=%4.3f out=%4.3f r=%d out->index(i=%d)=%d out->col_idxs()[i=%d]=%d first=%d last=%d gtid=%d\\n\",\n-// i, in->vals(0)[i], result, out->vals(0)[i],\n-// i / in->cols(), i, out->indexes()[i], i, out->col_idxs(0)[i], first_idx, last_idx, gtid);\n-// }\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/cuda/spoof/cellwise.cu", "new_path": "src/main/cuda/spoof/cellwise.cu", "diff": "@@ -53,8 +53,6 @@ struct SpoofCellwiseOp {\n}\n__device__ __forceinline__ T operator()(T a, uint32_t idx, uint32_t rix, uint32_t cix) {\n-//%NEED_RIX%\n-//%NEED_CIX%\n//%NEED_GRIX%\n%BODY_dense%\nreturn %OUT%;\n" } ]

[ { "change_type": "ADD", "old_path": null, "new_path": "scripts/builtin/hospitalResidencyMatch.dml", "diff": "+#-------------------------------------------------------------\n+## Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+##-------------------------------------------------------------\n+# THIS SCRIPT COMPUTES A SOLUTION FOR THE HOSPITAL RESIDENCY MATCH PROBLEM\n+#\n+# INPUT PARAMETERS:\n+# --------------------------------------------------------------------------------------------\n+# NAME TYPE DEFAULT MEANING\n+# --------------------------------------------------------------------------------------------\n+# R Matrix --- Residents matrix R.\n+# It must be an ORDERED matrix.\n+#\n+# H Matrix --- Hospitals matrix H.\n+# It must be an UNORDRED matrix.\n+#\n+# capacity Matrix --- capacity of Hospitals matrix C.\n+# It must be a [n*1] matrix with non zero values.\n+# i.e. the leftmost value in a row is the most preferred partner's index.\n+# i.e. the leftmost value in a row in P is the preference value for the acceptor\n+# with index 1 and vice-versa (higher is better).\n+# OUTPUT PARAMETERS:\n+# --------------------------------------------------------------------------------------------\n+# NAME TYPE DEFAULT MEANING\n+# --------------------------------------------------------------------------------------------\n+# residencyMatch Matrix --- Result Matrix\n+# If cell [i,j] is non-zero, it means that Resident i has matched with Hospital j.\n+# Further, if cell [i,j] is non-zero, it holds the preference value that led to the match.\n+#\n+#\n+# hospitalMatch Matrix --- Result Matrix\n+# If cell [i,j] is non-zero, it means that Resident i has matched with Hospital j.\n+# Further, if cell [i,j] is non-zero, it holds the preference value that led to the match.\n+#\n+#\n+# Residents.mtx:\n+# 2.0,1.0,3.0\n+# 1.0,2.0,3.0\n+# 1.0,2.0,0.0\n+#\n+# Since it is an ORDERED matrix, this means that Resident 1 (row 1) likes hospital 2 the most, followed by hospital 1 and hospital 3.\n+# If it was UNORDERED, this would mean that resident 1 (row 1) likes hospital 3 the most (since the value at [1,3] is the row max),\n+# followed by hospital 1 (2.0 preference value) and hospital 2 (1.0 preference value).\n+#\n+# Hospitals.mtx:\n+# 2.0,1.0,0.0\n+# 0.0,1.0,2.0\n+# 1.0,2.0,0.0\n+#\n+# Since it is an UNORDERED matrix this means that Hospital 1 (row 1) likes Resident 1 the most (since the value at [1,1] is the row max).\n+#\n+# capacity.mtx\n+# 1.0\n+# 1.0\n+# 1.0\n+#\n+# residencyMatch.mtx\n+# 2.0,0.0,0.0\n+# 1.0,0.0,0.0\n+# 0.0,2.0,0.0\n+#\n+\n+\n+# hospitalMatch.mtx\n+# 0.0,1.0,0.0\n+# 0.0,0.0,2.0\n+# 1.0,0.0,0.0\n+#\n+# Resident 1 has matched with Hospital 3 (since [1,3] is non-zero) at a preference level of 2.0.\n+# Resident 2 has matched with Hospital 1 (since [2,1] is non-zero) at a preference level of 1.0.\n+# Resident 3 has matched with Hospital 2 (since [3,2] is non-zero) at a preference level of 2.0.\n+# --------------------------------------------------------------------------------------------\n+\n+m_hospitalResidencyMatch = function(Matrix[Double] R, Matrix[Double] H, Matrix[Double] capacity, Boolean verbose = FALSE)\n+ return (Matrix[Double] residencyMatch, Matrix[Double] hospitalMatch)\n+{\n+\n+ # in this step we consider that Residents Matrix is ORDERED.\n+ # in this step we consider that Hospital Matrix is UNORDERED.\n+ # in the next implementation can consider number of choices for every resident.\n+\n+ #TODO set a finite number of maximum iterations so that the execution terminates after maximum iterations.\n+\n+ print(\"STARTING RESIDENCY MATCH ALGORITHM\");\n+ print(\"READING R as residents AND H as Hospitals and capacity...\");\n+\n+ m = nrow(R)\n+ n = ncol(R)\n+\n+ residencyMatch = matrix(0.0, rows=m, cols=n)\n+ hospitalMatch = matrix(0.0, rows=n, cols=m)\n+ resultmMatrix = matrix(0.0, rows=nrow(R), cols=ncol(R))\n+\n+ if(nrow(capacity) != nrow(H))\n+ print(\"ERROR: Missing capacity info for some hospitals\")\n+\n+\n+ startM = matrix(1.0, rows=m, cols=1) ### for checking while\n+\n+ hIndex =matrix(1.0, rows=m, cols=1)\n+ proposer_pointers = matrix(1.0, rows=m, cols=1)\n+ prev_Residents_vector = matrix(1.0, rows=n, cols=1)\n+ prevIndex_Residents_vector = matrix(1.0, rows=n, cols=1)\n+\n+ prev_Residents_vector = rowMins(hospitalMatch)\n+ prevIndex_Residents_vector = rowIndexMin(hospitalMatch)\n+ # TODO remove the nested looping by vectorizing\n+ while(sum(startM) > 0) {\n+ for(i in 1:m) {\n+ if(as.scalar(startM[i]) == 1) {\n+ secondIndex = as.scalar (proposer_pointers[i])\n+ hIndex[i] = as.scalar (R[i,secondIndex])\n+ #the minimum value means most preference.\n+ prev_Residents_vector = rowMaxs(hospitalMatch)\n+ prevIndex_Residents_vector = rowIndexMax(hospitalMatch)\n+ if (as.scalar(hIndex[i]) != 0) {\n+ hosValue = as.scalar (H[as.scalar(hIndex[i]),i])\n+ if (hosValue > 0) {\n+ # if this hospital likes this resident and has the capacity ...\n+ if(as.scalar(capacity[as.scalar (hIndex[i]),1]) >= 1) {\n+ capacity[as.scalar(hIndex[i]),1] = as.scalar(capacity[as.scalar (hIndex[i]),1]) - 1\n+ residencyMatch [i,as.scalar(hIndex[i])] = as.scalar(proposer_pointers[i])\n+ hospitalMatch [as.scalar(hIndex[i]), i] = hosValue\n+ #Disable freshly Matched resident to search for a new Hospital in the next round\n+ startM[i] = 0\n+ proposer_pointers[i] = as.scalar(proposer_pointers[i]) + 1\n+ if (as.scalar(proposer_pointers[i]) > n)\n+ proposer_pointers[i] = n\n+ }\n+ else if(as.scalar(prev_Residents_vector[as.scalar(hIndex[i])]) >= secondIndex) {\n+ #in this step we check that if the hospital capacity is 0\n+ # but the preference value of prev residents is lower than\n+ #the preference value of current resident.\n+ # we should replace the prev resident with current resident.\n+ resPrev= as.scalar(prevIndex_Residents_vector[as.scalar (hIndex[i]),1])\n+ hospitalMatch [as.scalar(hIndex[i]) ,resPrev] = 0\n+ residencyMatch[resPrev,as.scalar(hIndex[i])] = 0\n+ hospitalMatch [as.scalar(hIndex[i]),i ] = as.scalar(proposer_pointers[i])\n+ residencyMatch [i,as.scalar(hIndex[i])] = as.scalar(proposer_pointers[i])\n+ startM[i] = 0\n+ prevResIndex =as.scalar(prevIndex_Residents_vector[as.scalar(hIndex[i]),1])\n+ if(prevResIndex > 0){\n+ startM[prevResIndex ] =1\n+ proposer_pointers[i] = as.scalar(proposer_pointers[i]) + 1\n+ if (as.scalar(proposer_pointers[i]) > n)\n+ proposer_pointers[i] = n\n+ }\n+ }\n+ }\n+ if ( as.scalar (startM[i]) == 1 ) {\n+ proposer_pointers[i] = as.scalar(proposer_pointers[i]) + 1\n+ if (as.scalar(proposer_pointers[i]) > n)\n+ proposer_pointers[i] = n\n+ }\n+ }\n+ }\n+ }\n+ }\n+ if(verbose) {\n+ print(\"residencyMatch\")\n+ print(toString(residencyMatch))\n+ print(\"hospitalMatch\")\n+ print(toString(hospitalMatch))\n+ }\n+}\n+\n+\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/common/Builtins.java", "new_path": "src/main/java/org/apache/sysds/common/Builtins.java", "diff": "@@ -129,6 +129,7 @@ public enum Builtins {\nGMM_PREDICT(\"gmmPredict\", true),\nGNMF(\"gnmf\", true),\nGRID_SEARCH(\"gridSearch\", true),\n+ HOSPITAL_RESIDENCY_MATCH(\"hospitalResidencyMatch\", true),\nHYPERBAND(\"hyperband\", true),\nIFELSE(\"ifelse\", false),\nIMG_MIRROR(\"img_mirror\", true),\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysds/test/functions/builtin/BuiltinHospitalResidencyMatchTest.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysds.test.functions.builtin;\n+\n+import org.apache.sysds.common.Types;\n+import org.apache.sysds.runtime.matrix.data.MatrixValue;\n+import org.apache.sysds.test.AutomatedTestBase;\n+import org.apache.sysds.test.TestConfiguration;\n+import org.apache.sysds.test.TestUtils;\n+import org.junit.Test;\n+\n+import java.util.ArrayList;\n+import java.util.HashMap;\n+import java.util.List;\n+\n+public class BuiltinHospitalResidencyMatchTest extends AutomatedTestBase {\n+\n+\n+ private final static String TEST_NAME = \"residencymatch\";\n+ private final static String TEST_DIR = \"functions/builtin/\";\n+ private static final String TEST_CLASS_DIR = TEST_DIR + BuiltinHospitalResidencyMatchTest.class.getSimpleName() + \"/\";\n+\n+ private final static double eps = 0.0001;\n+\n+ @Override\n+ public void setUp() {\n+ addTestConfiguration(TEST_NAME,new TestConfiguration(TEST_CLASS_DIR, TEST_NAME,new String[]{\"RM\"}));\n+ }\n+\n+ @Test\n+ public void testResidencyMatch1() {\n+ double[][] R = {\n+ {2,3,1},{1,3,2},{3,1,3}};\n+ double[][] H = {\n+ {1,2,0},{3,1,2},{0,1,2}};\n+ double[][] C = {\n+ {2},{3},{2}};\n+ double[][]EM = { // this is an expected matrix\n+ {0,1,0},{1,0,0},{0,0,1}};\n+ runtestResidencyMatchTest(R, H, C, EM, Types.ExecType.CP);\n+ }\n+\n+ @Test\n+ public void testResidencyMatch2() {\n+ double[][] R = {\n+ {2,1,3},{1,2,3},{1,3,2}};\n+ double[][] H = {\n+ {3,1,2},{2,1,3},{3,2,1}};\n+ double[][] C = {\n+ {1},{1},{1}};\n+ double[][]EM = { // this is an expected matrix\n+ {0,0,3},{0,2,0},{1,0,0}};\n+ runtestResidencyMatchTest(R, H, C, EM, Types.ExecType.CP);\n+ }\n+\n+ @Test\n+ public void testResidencyMatch3() {\n+ double[][] R = {\n+ {1,2},{2,1},{1,2},{1,2}};\n+ double[][] H = {\n+ {3,2,1,4},{2,1,3,0}};\n+ double[][] C = {\n+ {4},{3}};\n+ double[][]EM = { // this is an expected matrix\n+ {1,0},{0,1},{1,0},{1,0}};\n+ runtestResidencyMatchTest(R, H, C, EM, Types.ExecType.CP);\n+ }\n+ @Test\n+ public void testResidencyMatch4() {\n+ double[][] R = {\n+ {1,2},{2,1},{1,2},{1,2}};\n+ double[][] H = {\n+ {3,2,1,4},{2,1,3,0}};\n+ double[][] C = {\n+ {4},{3}};\n+ double[][]EM = { // this is an expected matrix\n+ {1,0},{0,1},{1,0},{1,0}};\n+ runtestResidencyMatchTest(R, H, C, EM, Types.ExecType.SPARK);\n+ }\n+\n+ private void runtestResidencyMatchTest(double[][] R, double[][] H, double[][] C, double[][] EM,\n+ Types.ExecType instType) {\n+\n+ Types.ExecMode platformOld = setExecMode(instType);\n+ try {\n+ loadTestConfiguration(getTestConfiguration(TEST_NAME));\n+ String HOME = SCRIPT_DIR + TEST_DIR;\n+ fullDMLScriptName = HOME + TEST_NAME + \".dml\";\n+ List<String> proArgs = new ArrayList<>();\n+ proArgs.add(\"-args\");\n+ proArgs.add(input(\"R\"));\n+ proArgs.add(input(\"H\"));\n+ proArgs.add(input(\"C\"));\n+ proArgs.add(output(\"RM\"));\n+\n+ programArgs = proArgs.toArray(new String[proArgs.size()]);\n+ // defining Residents Matrix\n+\n+ writeInputMatrixWithMTD(\"R\", R, true);\n+ writeInputMatrixWithMTD(\"H\", H, true);\n+ writeInputMatrixWithMTD(\"C\", C, true);\n+\n+\n+ runTest(true, EXCEPTION_NOT_EXPECTED, null, -1);\n+\n+ //compare expected results\n+ HashMap<MatrixValue.CellIndex, Double> matrixU = readDMLMatrixFromOutputDir(\"RM\");\n+ double[][] OUT = TestUtils.convertHashMapToDoubleArray(matrixU);\n+ TestUtils.compareMatrices(EM, OUT, eps);\n+ }\n+ finally {\n+ rtplatform = platformOld;\n+ }\n+ }\n+}\n+\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/builtin/residencymatch.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+R = read($1)\n+H = read($2)\n+C = read($3)\n+[R, H] = hospitalResidencyMatch(R=R,H=H,capacity=C, verbose = TRUE)\n+write(R, $4)\n+\n+\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "<excludes>\n<exclude>algorithms/obsolete/*</exclude>\n<exclude>datagen/obsolete/*</exclude>\n- <exclude>perftest/*</exclude>\n+ <exclude>perftest/**/*</exclude>\n+ <exclude>perftest</exclude>\n+ <exclude>perftestDeprecated/*</exclude>\n+ <exclude>perftestDeprecated</exclude>\n<exclude>staging/**/*</exclude>\n<exclude>nn/test/compare_backends/*</exclude>\n<exclude>nn/test/compare_backends/*</exclude>\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/cuda/headers/agg_ops.cuh", "new_path": "src/main/cuda/headers/agg_ops.cuh", "diff": "@@ -92,8 +92,6 @@ struct IdentityOp {\ntemplate<typename T>\nstruct SumOp {\n__device__ __forceinline__ T operator()(T a, T b) const {\n-// if(blockIdx.x==0 && threadIdx.x ==0)\n-// printf(\"a=%f + b=%f => %f\\n\", a, b, a+b);\nreturn a + b;\n}\n@@ -124,16 +122,6 @@ struct MinusOp {\n}\n};\n-/**\n- * Functor op for sum of squares operation (returns a + b * b)\n- */\n-template<typename T>\n-struct SumSqOp {\n- __device__ __forceinline__ T operator()(T a, T b) const {\n- return a + b * b;\n- }\n-};\n-\n/**\n* Functor op for min operation\n*/\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/codegen/cplan/CNodeCell.java", "new_path": "src/main/java/org/apache/sysds/hops/codegen/cplan/CNodeCell.java", "diff": "@@ -162,8 +162,9 @@ public class CNodeCell extends CNodeTpl\ntmp = tmp.replace(\"%BODY_dense%\", tmpDense);\n- //return last TMP\n- tmp = tmp.replaceAll(\"%OUT%\", _output.getVarname());\n+ //Return last TMP. Square it for CUDA+SUM_SQ\n+ tmp = (api.isJava() || _aggOp != AggOp.SUM_SQ) ? tmp.replaceAll(\"%OUT%\", _output.getVarname()) :\n+ tmp.replaceAll(\"%OUT%\", _output.getVarname() + \" * \" + _output.getVarname());\n//replace meta data information\ntmp = tmp.replaceAll(\"%TYPE%\", getCellType().name());\n@@ -181,11 +182,8 @@ public class CNodeCell extends CNodeTpl\nif(_aggOp != null)\nswitch(_aggOp) {\ncase SUM:\n- agg_op = \"SumOp\";\n- initial_value = \"(T)0.0\";\n- break;\ncase SUM_SQ:\n- agg_op = \"SumSqOp\";\n+ agg_op = \"SumOp\";\ninitial_value = \"(T)0.0\";\nbreak;\ncase MIN:\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/codegen/SystemDS-config-codegen6.xml", "new_path": "src/test/scripts/functions/codegen/SystemDS-config-codegen6.xml", "diff": "<!-- The number of theads for the spark instance artificially selected-->\n<sysds.local.spark.number.threads>16</sysds.local.spark.number.threads>\n+\n+ <sysds.codegen.api>auto</sysds.codegen.api>\n</root>\n\\ No newline at end of file\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/gpu/SpoofCUDAInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/gpu/SpoofCUDAInstruction.java", "diff": "@@ -38,6 +38,7 @@ import org.apache.sysds.runtime.instructions.gpu.context.GPUObject;\nimport org.apache.sysds.runtime.lineage.LineageItem;\nimport org.apache.sysds.runtime.lineage.LineageItemUtils;\nimport org.apache.sysds.runtime.lineage.LineageTraceable;\n+import org.apache.sysds.utils.GPUStatistics;\nimport java.util.ArrayList;\n@@ -116,6 +117,7 @@ public class SpoofCUDAInstruction extends GPUInstruction implements LineageTrace\n@Override\npublic void processInstruction(ExecutionContext ec) {\n+ GPUStatistics.incrementNoOfExecutedGPUInst();\n//get input matrices and scalars, incl pinning of matrices\nArrayList<MatrixObject> inputs = new ArrayList<>();\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "<version>18</version>\n</parent>\n<groupId>org.apache.systemds</groupId>\n- <version>2.2.0-SNAPSHOT</version>\n+ <version>2.1.0</version>\n<artifactId>systemds</artifactId>\n<packaging>jar</packaging>\n<name>SystemDS</name>\n<scm>\n<developerConnection>scm:git:https://github.com/apache/systemds.git</developerConnection>\n- <tag>HEAD</tag>\n+ <tag>2.1.0-rc3</tag>\n</scm>\n<build>\n" } ]

[ { "change_type": "MODIFY", "old_path": "README.md", "new_path": "README.md", "diff": "@@ -37,7 +37,7 @@ heterogeneous and nested schema.\n**Issue Tracker** [Jira Dashboard](https://issues.apache.org/jira/secure/Dashboard.jspa?selectPageId=12335852)\n**Status and Build:** SystemDS is renamed from SystemML which is an **Apache Top Level Project**.\n-To build from source visit [SystemDS Install from source](http://apache.github.io/systemds/site/install.html)\n+To build from source visit [SystemDS Install from source](https://apache.github.io/systemds/site/install.html)\n[](https://github.com/apache/systemds/actions?query=workflow%3A%22Build%22+branch%3Amaster+event%3Apush)\n[](https://github.com/apache/systemds/actions?query=workflow%3ADocumentation+branch%3Amaster+event%3Apush)\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/systemds/project_info.py", "new_path": "src/main/python/systemds/project_info.py", "diff": "# via string substitutions using the maven-resources-plugin\n__project_group_id__ = 'org.apache.systemds'\n__project_artifact_id__ = 'systemds'\n-__project_version__ = '2.1.0-SNAPSHOT'\n+__project_version__ = '2.2.0-SNAPSHOT'\n" } ]

[ { "change_type": "MODIFY", "old_path": "docs/_config.yml", "new_path": "docs/_config.yml", "diff": "@@ -39,7 +39,7 @@ exclude:\n- updateAPI.sh\n# These allow the documentation to be updated with newer releases\n-SYSTEMDS_VERSION: 2.1.0-SNAPSHOT\n+SYSTEMDS_VERSION: 2.2.0-SNAPSHOT\n# if 'analytics_on' is true, analytics section will be rendered on the HTML pages\nanalytics_on: true\n" } ]

[ { "change_type": "MODIFY", "old_path": "docs/updateAPI.sh", "new_path": "docs/updateAPI.sh", "diff": "@@ -39,7 +39,7 @@ else\nrm -r api/python\nmkdir api/python\ncd ../src/main/python/docs\n- rm -r build\n+ rm -f -r build\nmake html\ncd ../../../../\ncp -r src/main/python/docs/build/html/* docs/api/python\n" } ]

[ { "change_type": "MODIFY", "old_path": "CONTRIBUTING.md", "new_path": "CONTRIBUTING.md", "diff": "@@ -92,9 +92,12 @@ The tags can be used in combination to one another. These are the only tags avai\n> [`87bc3584`](https://github.com/apache/systemds/commit/87bc3584db2148cf78b2d46418639e88ca27ec64) - `[HOTFIX] Fix validation of scalar-scalar binary min/max operations`\n>\n+> Protip:\n+> Addressing multiple jira issues in a single commit, `[SYSTEMDS-123,SYSTEMDS-124]` or `[SYSTEMDS-123][SYSTEMDS-124]`\n+\n### Commit description\n-> A commit or PR description is a public record of **what** change is being made and **why**.\n+A commit or PR description is a public record of **what** change is being made and **why**.\n#### Structure of the description\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/docs/source/api/operator/algorithms.rst", "new_path": "src/main/python/docs/source/api/operator/algorithms.rst", "diff": "@@ -28,7 +28,7 @@ As an example the lm algorithm can be used as follows:\n.. code-block:: python\n- # Import numpy and SystemDS matrix\n+ # Import numpy and SystemDS\nimport numpy as np\nfrom systemds.context import SystemDSContext\nfrom systemds.operator.algorithm import lm\n" }, { "change_type": "MODIFY", "old_path": "src/main/python/docs/source/getting_started/simple_examples.rst", "new_path": "src/main/python/docs/source/getting_started/simple_examples.rst", "diff": "@@ -90,7 +90,7 @@ One example of this is l2SVM, a high level functions for Data-Scientists. Let's\n.. code-block:: python\n- # Import numpy and SystemDS matrix\n+ # Import numpy and SystemDS\nimport numpy as np\nfrom systemds.context import SystemDSContext\nfrom systemds.operator.algorithm import l2svm\n" }, { "change_type": "MODIFY", "old_path": "src/main/python/docs/source/guide/algorithms_basics.rst", "new_path": "src/main/python/docs/source/guide/algorithms_basics.rst", "diff": "@@ -96,7 +96,6 @@ Then setup the data\n.. code-block:: python\n- from systemds.operator import Matrix\nX_ds = sds.from_numpy(X)\nY_ds = sds.from_numpy( Y)\n@@ -210,12 +209,12 @@ this makes SystemDS responsible for adding the 1 to each value.\nwith SystemDSContext() as sds:\n# Train Data\n- X = Matrix(sds, d.get_train_data().reshape((60000, 28*28)))\n- Y = Matrix(sds, d.get_train_labels()) + 1.0\n+ X = sds.from_numpy(d.get_train_data().reshape((60000, 28*28)))\n+ Y = sds.from_numpy(d.get_train_labels()) + 1.0\nbias = multiLogReg(X, Y, maxi=30)\n# Test data\n- Xt = Matrix(sds, d.get_test_data().reshape((10000, 28*28)))\n- Yt = Matrix(sds, d.get_test_labels()) + 1.0\n+ Xt = sds.from_numpy(d.get_test_data().reshape((10000, 28*28)))\n+ Yt = sds.from_numpy(d.get_test_labels()) + 1.0\n[m, y_pred, acc] = multiLogRegPredict(Xt, bias, Yt).compute()\nprint(acc)\n" }, { "change_type": "MODIFY", "old_path": "src/main/python/docs/source/guide/federated.rst", "new_path": "src/main/python/docs/source/guide/federated.rst", "diff": "@@ -67,9 +67,8 @@ The aggregated sum using federated instructions in python SystemDS is done as fo\n.. code-block:: python\n- # Import numpy and SystemDS federated\n+ # Import numpy and SystemDS\nimport numpy as np\n- from systemds.matrix import Federated\nfrom systemds.context import SystemDSContext\n# Create a federated matrix\n@@ -83,7 +82,7 @@ The aggregated sum using federated instructions in python SystemDS is done as fo\naddress = \"localhost:8001/temp/test.csv\"\nwith SystemDSContext() as sds:\n- fed_a = Federated(sds, [address], [dims])\n+ fed_a = sds.federated(sds, [address], [dims])\n# Sum the federated matrix and call compute to execute\nprint(fed_a.sum().compute())\n# Result should be 45.\n@@ -107,7 +106,6 @@ Once all three workers are up and running we can leverage all three in the follo\n.. code-block:: python\n- # Import numpy and SystemDS federated\nimport numpy as np\nfrom systemds.context import SystemDSContext\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/systemds/operator/algorithm/__init__.py", "new_path": "src/main/python/systemds/operator/algorithm/__init__.py", "diff": "@@ -48,6 +48,7 @@ from .builtin.glm import glm\nfrom .builtin.gmm import gmm\nfrom .builtin.gmmPredict import gmmPredict\nfrom .builtin.gnmf import gnmf\n+from .builtin.hospitalResidencyMatch import hospitalResidencyMatch\nfrom .builtin.hyperband import hyperband\nfrom .builtin.img_brightness import img_brightness\nfrom .builtin.img_crop import img_crop\n@@ -100,6 +101,7 @@ from .builtin.sherlockPredict import sherlockPredict\nfrom .builtin.sigmoid import sigmoid\nfrom .builtin.slicefinder import slicefinder\nfrom .builtin.smote import smote\n+from .builtin.softmax import softmax\nfrom .builtin.split import split\nfrom .builtin.splitBalanced import splitBalanced\nfrom .builtin.stableMarriage import stableMarriage\n@@ -140,6 +142,7 @@ __all__ = ['abstain',\n'gmm',\n'gmmPredict',\n'gnmf',\n+ 'hospitalResidencyMatch',\n'hyperband',\n'img_brightness',\n'img_crop',\n@@ -192,6 +195,7 @@ __all__ = ['abstain',\n'sigmoid',\n'slicefinder',\n'smote',\n+ 'softmax',\n'split',\n'splitBalanced',\n'stableMarriage',\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/python/systemds/operator/algorithm/builtin/hospitalResidencyMatch.py", "diff": "+# -------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+# -------------------------------------------------------------\n+\n+# Autogenerated By : src/main/python/generator/generator.py\n+# Autogenerated From : scripts/builtin/hospitalResidencyMatch.dml\n+\n+from typing import Dict, Iterable\n+\n+from systemds.operator import OperationNode, Matrix, Frame, List, MultiReturn, Scalar\n+from systemds.script_building.dag import OutputType\n+from systemds.utils.consts import VALID_INPUT_TYPES\n+\n+\n+def hospitalResidencyMatch(R: Matrix,\n+ H: Matrix,\n+ capacity: Matrix,\n+ **kwargs: Dict[str, VALID_INPUT_TYPES]):\n+ \"\"\"\n+ :param R: Residents matrix R.\n+ :param It: an ORDERED matrix.\n+ :param H: Hospitals matrix H.\n+ :param It: an UNORDRED matrix.\n+ :param capacity: capacity of Hospitals matrix C.\n+ :param It: a [n*1] matrix with non zero values.\n+ :param with: and vice-versa (higher is better).\n+ :return: 'OperationNode' containing result matrix & result matrix & an ordered matrix, this means that resident 1 (row 1) likes hospital 2 the most, followed by hospital 1 and hospital 3. & unordered, this would mean that resident 1 (row 1) likes hospital 3 the most (since the value at [1,3] is the row max), & 1 (2.0 preference value) and hospital 2 (1.0 preference value). & an unordered matrix this means that hospital 1 (row 1) likes resident 1 the most (since the value at [1,1] is the row max). & matched with hospital 3 (since [1,3] is non-zero) at a preference level of 2.0. & matched with hospital 1 (since [2,1] is non-zero) at a preference level of 1.0. & matched with hospital 2 (since [3,2] is non-zero) at a preference level of 2.0.\n+ \"\"\"\n+ params_dict = {'R': R, 'H': H, 'capacity': capacity}\n+ params_dict.update(kwargs)\n+\n+ vX_0 = Matrix(R.sds_context, '')\n+ vX_1 = Matrix(R.sds_context, '')\n+ output_nodes = [vX_0, vX_1, ]\n+\n+ op = MultiReturn(R.sds_context, 'hospitalResidencyMatch', output_nodes, named_input_nodes=params_dict)\n+\n+ vX_0._unnamed_input_nodes = [op]\n+ vX_1._unnamed_input_nodes = [op]\n+\n+ return op\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/python/systemds/operator/algorithm/builtin/softmax.py", "diff": "+# -------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+# -------------------------------------------------------------\n+\n+# Autogenerated By : src/main/python/generator/generator.py\n+# Autogenerated From : scripts/builtin/softmax.dml\n+\n+from typing import Dict, Iterable\n+\n+from systemds.operator import OperationNode, Matrix, Frame, List, MultiReturn, Scalar\n+from systemds.script_building.dag import OutputType\n+from systemds.utils.consts import VALID_INPUT_TYPES\n+\n+\n+def softmax(S: Matrix):\n+\n+ params_dict = {'S': S}\n+ return Matrix(S.sds_context,\n+ 'softmax',\n+ named_input_nodes=params_dict)\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/python/tests/matrix/test_split.py", "diff": "+# -------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+# -------------------------------------------------------------\n+\n+import unittest\n+import random\n+\n+import numpy as np\n+from systemds.context import SystemDSContext\n+from systemds.operator.algorithm import split\n+\n+# Seed the random ness.\n+np.random.seed(7)\n+\n+class TestOrder(unittest.TestCase):\n+\n+ sds: SystemDSContext = None\n+\n+ @classmethod\n+ def setUpClass(cls):\n+ cls.sds = SystemDSContext()\n+\n+ @classmethod\n+ def tearDownClass(cls):\n+ cls.sds.close()\n+\n+ def test_basic(self):\n+ m = self.make_matrix()\n+\n+ o = self.sds.from_numpy(m).compute()\n+ s = m\n+ self.assertTrue(np.allclose(o, s))\n+\n+ def test_split(self):\n+ X = self.make_matrix()\n+ Y = self.make_matrix(cols = 2)\n+\n+ [p1,p2,p3,p4] = split(self.sds.from_numpy(X), self.sds.from_numpy(Y)).compute()\n+ exp1 = X[:2]\n+ exp2 = X[2:]\n+ exp3 = Y[:2]\n+ exp4 = Y[2:]\n+ self.assertTrue(np.allclose(p1, exp1))\n+ self.assertTrue(np.allclose(p2, exp2))\n+ self.assertTrue(np.allclose(p3, exp3))\n+ self.assertTrue(np.allclose(p4, exp4))\n+\n+ def test_split_2(self):\n+ rows = 10\n+ X = self.make_matrix(rows = rows)\n+ Y = self.make_matrix(rows = rows, cols = 2)\n+\n+ [p1,p2,p3,p4] = split(self.sds.from_numpy(X), self.sds.from_numpy(Y)).compute()\n+ exp1 = X[:7]\n+ exp2 = X[7:]\n+ exp3 = Y[:7]\n+ exp4 = Y[7:]\n+ self.assertTrue(np.allclose(p1, exp1))\n+ self.assertTrue(np.allclose(p2, exp2))\n+ self.assertTrue(np.allclose(p3, exp3))\n+ self.assertTrue(np.allclose(p4, exp4))\n+\n+ def test_split_3(self):\n+ rows = 100\n+ X = self.make_matrix(rows = rows)\n+ Y = self.make_matrix(rows = rows, cols = 2)\n+\n+ [p1,p2,p3,p4] = split(self.sds.from_numpy(X), self.sds.from_numpy(Y)).compute()\n+ exp1 = X[:70]\n+ exp2 = X[70:]\n+ exp3 = Y[:70]\n+ exp4 = Y[70:]\n+ self.assertTrue(np.allclose(p1, exp1))\n+ self.assertTrue(np.allclose(p2, exp2))\n+ self.assertTrue(np.allclose(p3, exp3))\n+ self.assertTrue(np.allclose(p4, exp4))\n+\n+\n+ def test_split_4(self):\n+ rows = 100\n+ X = self.make_matrix(rows = rows)\n+ Y = self.make_matrix(rows = rows, cols = 2)\n+\n+ [p1,p2,p3,p4] = split(self.sds.from_numpy(X), self.sds.from_numpy(Y), f=0.2).compute()\n+ exp1 = X[:20]\n+ exp2 = X[20:]\n+ exp3 = Y[:20]\n+ exp4 = Y[20:]\n+ self.assertTrue(np.allclose(p1, exp1))\n+ self.assertTrue(np.allclose(p2, exp2))\n+ self.assertTrue(np.allclose(p3, exp3))\n+ self.assertTrue(np.allclose(p4, exp4))\n+\n+\n+ def make_matrix(self, rows = 4, cols = 4):\n+ return np.random.rand(rows,cols)\n+\n+\n+if __name__ == \"__main__\":\n+ unittest.main(exit=False)\n" } ]

[ { "change_type": "MODIFY", "old_path": "docs/updateAPI.sh", "new_path": "docs/updateAPI.sh", "diff": "@@ -49,7 +49,7 @@ else\nfind . -type f -exec sed -i 's/_static/static/g' {} +\nmv _sources sources\nfind . -type f -exec sed -i 's/_sources/sources/g' {} +\n- mv _images images\n+ [[ -f \"_images\" ]] && mv _images images\nfind . -type f -exec sed -i 's/_images/images/g' {} +\ncd ../../\nfi\n" } ]

[ { "change_type": "MODIFY", "old_path": "docs/updateAPI.sh", "new_path": "docs/updateAPI.sh", "diff": "#\n#-------------------------------------------------------------\n+# stop the script at first error\n+set -e\n+\ncurFolder=${PWD##*/}\nif [ $curFolder != \"docs\" ]; then\n@@ -49,7 +52,15 @@ else\nfind . -type f -exec sed -i 's/_static/static/g' {} +\nmv _sources sources\nfind . -type f -exec sed -i 's/_sources/sources/g' {} +\n- [[ -f \"_images\" ]] && mv _images images\n+\n+ if [[ -d \"_images\" ]]\n+ then\n+ mv _images images\nfind . -type f -exec sed -i 's/_images/images/g' {} +\n+ fi\n+\ncd ../../\nfi\n+\n+exit 0\n+\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/Hop.java", "new_path": "src/main/java/org/apache/sysds/hops/Hop.java", "diff": "package org.apache.sysds.hops;\n+import java.util.ArrayList;\n+import java.util.Collection;\n+import java.util.HashMap;\n+import java.util.HashSet;\n+\nimport org.apache.commons.logging.Log;\nimport org.apache.commons.logging.LogFactory;\nimport org.apache.sysds.api.DMLScript;\nimport org.apache.sysds.common.Types.DataType;\nimport org.apache.sysds.common.Types.ExecMode;\n+import org.apache.sysds.common.Types.ExecType;\nimport org.apache.sysds.common.Types.FileFormat;\nimport org.apache.sysds.common.Types.OpOp1;\nimport org.apache.sysds.common.Types.OpOp2;\n@@ -38,12 +44,13 @@ import org.apache.sysds.lops.Compression;\nimport org.apache.sysds.lops.Data;\nimport org.apache.sysds.lops.DeCompression;\nimport org.apache.sysds.lops.Lop;\n-import org.apache.sysds.common.Types.ExecType;\nimport org.apache.sysds.lops.LopsException;\nimport org.apache.sysds.lops.ReBlock;\nimport org.apache.sysds.lops.UnaryCP;\nimport org.apache.sysds.parser.ParseInfo;\n+import org.apache.sysds.runtime.compress.workload.AWTreeNode;\nimport org.apache.sysds.runtime.controlprogram.LocalVariableMap;\n+import org.apache.sysds.runtime.controlprogram.SingletonLookupHashMap;\nimport org.apache.sysds.runtime.controlprogram.caching.MatrixObject.UpdateType;\nimport org.apache.sysds.runtime.controlprogram.context.SparkExecutionContext;\nimport org.apache.sysds.runtime.controlprogram.parfor.util.IDSequence;\n@@ -55,11 +62,6 @@ import org.apache.sysds.runtime.meta.MatrixCharacteristics;\nimport org.apache.sysds.runtime.privacy.PrivacyConstraint;\nimport org.apache.sysds.runtime.util.UtilFunctions;\n-import java.util.ArrayList;\n-import java.util.Collection;\n-import java.util.HashMap;\n-import java.util.HashSet;\n-\npublic abstract class Hop implements ParseInfo {\nprivate static final Log LOG = LogFactory.getLog(Hop.class.getName());\n@@ -107,10 +109,17 @@ public abstract class Hop implements ParseInfo {\n// (usually this happens on persistent reads dataops)\nprotected boolean _requiresReblock = false;\n- // indicates if the output of this hop needs to be compressed\n- // (this happens on persistent reads after reblock but before checkpoint)\n+ /**\n+ * indicates if the output of this hop needs to be compressed\n+ * (this happens on persistent reads after reblock but before checkpoint)\n+ */\nprotected boolean _requiresCompression = false;\n+ /**\n+ * A WTree for this hop instruction in case the compression\n+ */\n+ protected AWTreeNode _compressedWorkloadTree = null;\n+\n/** Boolean specifying if decompression is required.*/\nprotected boolean _requiresDeCompression = false;\n@@ -268,12 +277,17 @@ public abstract class Hop implements ParseInfo {\nreturn _requiresCheckpoint;\n}\n- public void setRequiresCompression(boolean flag) {\n- _requiresCompression = flag;\n+ public void setRequiresCompression(){\n+ _requiresCompression = true;\n}\n- public void setRequiresDeCompression(boolean flag){\n- _requiresDeCompression = flag;\n+ public void setRequiresCompression(AWTreeNode node) {\n+ _requiresCompression = true;\n+ _compressedWorkloadTree = node;\n+ }\n+\n+ public void setRequiresDeCompression(){\n+ _requiresDeCompression = true;\n}\npublic boolean requiresCompression() {\n@@ -394,8 +408,15 @@ public abstract class Hop implements ParseInfo {\nExecType et = getExecutionModeForCompression();\nLop compressionInstruction = null;\ntry{\n- if( _requiresCompression )\n- compressionInstruction = new Compression(getLops(), getDataType(), getValueType(), et);\n+ if( _requiresCompression ){\n+ if(_compressedWorkloadTree != null){\n+ int singletonID = SingletonLookupHashMap.getMap().put(_compressedWorkloadTree);\n+ compressionInstruction = new Compression(getLops(), getDataType(), getValueType(), et, singletonID);\n+ }\n+ else {\n+ compressionInstruction = new Compression(getLops(), getDataType(), getValueType(), et, 0);\n+ }\n+ }\nelse if( _requiresDeCompression )\ncompressionInstruction = new DeCompression(getLops(), getDataType(), getValueType(), et);\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/OptimizerUtils.java", "new_path": "src/main/java/org/apache/sysds/hops/OptimizerUtils.java", "diff": "@@ -226,6 +226,12 @@ public class OptimizerUtils\n*/\npublic static final boolean ALLOW_SCRIPT_LEVEL_COMPRESS_COMMAND = true;\n+ /**\n+ * Boolean specifying if compression rewrites is allowed. This is disabled at run time if the IPA for Workload aware compression\n+ * is activated.\n+ */\n+ public static boolean ALLOW_COMPRESSION_REWRITE = true;\n+\n//////////////////////\n// Optimizer levels //\n//////////////////////\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/ipa/IPAPassCompressionWorkloadAnalysis.java", "new_path": "src/main/java/org/apache/sysds/hops/ipa/IPAPassCompressionWorkloadAnalysis.java", "diff": "@@ -24,34 +24,46 @@ import java.util.Map.Entry;\nimport org.apache.sysds.conf.ConfigurationManager;\nimport org.apache.sysds.conf.DMLConfig;\n+import org.apache.sysds.hops.OptimizerUtils;\nimport org.apache.sysds.lops.Compression.CompressConfig;\nimport org.apache.sysds.parser.DMLProgram;\n-import org.apache.sysds.runtime.compress.workload.WTreeNode;\n+import org.apache.sysds.runtime.compress.workload.WTreeRoot;\nimport org.apache.sysds.runtime.compress.workload.WorkloadAnalyzer;\n/**\n- * This rewrite obtains workload summaries for all hops candidates amenable\n- * for compression as a basis for workload-aware compression planning.\n- *\n+ * This rewrite obtains workload summaries for all hops candidates amenable for compression as a basis for\n+ * workload-aware compression planning.\n*/\n-public class IPAPassCompressionWorkloadAnalysis extends IPAPass\n-{\n+public class IPAPassCompressionWorkloadAnalysis extends IPAPass {\n+\n@Override\npublic boolean isApplicable(FunctionCallGraph fgraph) {\n- return InterProceduralAnalysis.CLA_WORKLOAD_ANALYSIS\n- && CompressConfig.valueOf(ConfigurationManager.getDMLConfig()\n- .getTextValue(DMLConfig.COMPRESSED_LINALG).toUpperCase()).isEnabled();\n+ return InterProceduralAnalysis.CLA_WORKLOAD_ANALYSIS && CompressConfig\n+ .valueOf(ConfigurationManager.getDMLConfig().getTextValue(DMLConfig.COMPRESSED_LINALG).toUpperCase())\n+ .isEnabled();\n}\n@Override\npublic boolean rewriteProgram(DMLProgram prog, FunctionCallGraph fgraph, FunctionCallSizeInfo fcallSizes) {\n- //obtain CLA workload analysis for all applicable operators\n- Map<Long, WTreeNode> map = WorkloadAnalyzer.getAllCandidateWorkloads(prog);\n+ // Parse compression config\n+ DMLConfig conf = ConfigurationManager.getDMLConfig();\n+ CompressConfig compress = CompressConfig.valueOf(conf.getTextValue(DMLConfig.COMPRESSED_LINALG).toUpperCase());\n+ // Verify that we have Workload enabled.\n+ if(compress == CompressConfig.WORKLOAD) {\n+ // Set rewrite rule for CLA to false, since we are using workload based planning.\n+ OptimizerUtils.ALLOW_COMPRESSION_REWRITE = false;\n+\n+ // Obtain CLA workload analysis for all applicable operators\n+ Map<Long, WTreeRoot> map = WorkloadAnalyzer.getAllCandidateWorkloads(prog);\n- //TODO influence compression planning, for now just printing\n- for( Entry<Long, WTreeNode> e : map.entrySet() )\n- System.out.println(e.getValue());\n+ // TODO Prune away obviously bad compression locations.\n+\n+ // Add compression instruction to all remaining locations\n+ for(Entry<Long, WTreeRoot> e : map.entrySet())\n+ e.getValue().getRoot().setRequiresCompression(e.getValue());\nreturn map != null;\n}\n+ return false;\n+ }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/rewrite/ProgramRewriter.java", "new_path": "src/main/java/org/apache/sysds/hops/rewrite/ProgramRewriter.java", "diff": "@@ -107,6 +107,7 @@ public class ProgramRewriter\n_sbRuleSet.add( new RewriteRemoveForLoopEmptySequence() ); //dependency: constant folding\nif( OptimizerUtils.ALLOW_BRANCH_REMOVAL || OptimizerUtils.ALLOW_FOR_LOOP_REMOVAL )\n_sbRuleSet.add( new RewriteMergeBlockSequence() ); //dependency: remove branches, remove for-loops\n+ if(OptimizerUtils.ALLOW_COMPRESSION_REWRITE)\n_sbRuleSet.add( new RewriteCompressedReblock() ); // Compression Rewrite\nif( OptimizerUtils.ALLOW_SPLIT_HOP_DAGS )\n_sbRuleSet.add( new RewriteSplitDagUnknownCSVRead() ); //dependency: reblock, merge blocks\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/rewrite/RewriteCompressedReblock.java", "new_path": "src/main/java/org/apache/sysds/hops/rewrite/RewriteCompressedReblock.java", "diff": "@@ -105,22 +105,22 @@ public class RewriteCompressedReblock extends StatementBlockRewriteRule {\nswitch(compress) {\ncase TRUE:\nif(satisfiesCompressionCondition(hop))\n- hop.setRequiresCompression(true);\n+ hop.setRequiresCompression();\nbreak;\ncase AUTO:\nif(OptimizerUtils.isSparkExecutionMode() && satisfiesAutoCompressionCondition(hop, prog))\n- hop.setRequiresCompression(true);\n+ hop.setRequiresCompression();\nbreak;\ncase COST:\nif(satisfiesCostCompressionCondition(hop, prog))\n- hop.setRequiresCompression(true);\n+ hop.setRequiresCompression();\nbreak;\ndefault:\nbreak;\n}\nif(satisfiesDeCompressionCondition(hop)) {\n- hop.setRequiresDeCompression(true);\n+ hop.setRequiresDeCompression();\n}\nhop.setVisited();\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/lops/Compression.java", "new_path": "src/main/java/org/apache/sysds/lops/Compression.java", "diff": "package org.apache.sysds.lops;\n-import org.apache.sysds.common.Types.ExecType;\nimport org.apache.sysds.common.Types.DataType;\n+import org.apache.sysds.common.Types.ExecType;\nimport org.apache.sysds.common.Types.ValueType;\npublic class Compression extends Lop {\npublic static final String OPCODE = \"compress\";\n+ private final int _singletonLookupKey;\n+\npublic enum CompressConfig {\n- TRUE, FALSE, COST, AUTO;\n+ TRUE, FALSE, COST, AUTO, WORKLOAD;\npublic boolean isEnabled() {\nreturn this != FALSE;\n}\n}\n- public Compression(Lop input, DataType dt, ValueType vt, ExecType et) {\n+ public Compression(Lop input, DataType dt, ValueType vt, ExecType et, int singletonLookupKey) {\nsuper(Lop.Type.Checkpoint, dt, vt);\naddInput(input);\ninput.addOutput(this);\nlps.setProperties(inputs, et);\n+ _singletonLookupKey = singletonLookupKey;\n}\n@Override\n@@ -56,6 +59,11 @@ public class Compression extends Lop {\nsb.append(getInputs().get(0).prepInputOperand(input1));\nsb.append(OPERAND_DELIMITOR);\nsb.append(prepOutputOperand(output));\n+ if(_singletonLookupKey != 0){\n+ sb.append(OPERAND_DELIMITOR);\n+ sb.append(_singletonLookupKey);\n+ }\n+\nreturn sb.toString();\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/CompressedMatrixBlockFactory.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/CompressedMatrixBlockFactory.java", "diff": "@@ -39,6 +39,7 @@ import org.apache.sysds.runtime.compress.estim.CompressedSizeEstimator;\nimport org.apache.sysds.runtime.compress.estim.CompressedSizeEstimatorFactory;\nimport org.apache.sysds.runtime.compress.estim.CompressedSizeInfo;\nimport org.apache.sysds.runtime.compress.utils.DblArrayIntListHashMap;\n+import org.apache.sysds.runtime.compress.workload.WTreeRoot;\nimport org.apache.sysds.runtime.controlprogram.parfor.stat.Timing;\nimport org.apache.sysds.runtime.matrix.data.LibMatrixReorg;\nimport org.apache.sysds.runtime.matrix.data.MatrixBlock;\n@@ -88,6 +89,10 @@ public class CompressedMatrixBlockFactory {\nreturn compress(mb, k, new CompressionSettingsBuilder().create());\n}\n+ public static Pair<MatrixBlock, CompressionStatistics> compress(MatrixBlock mb, int k, WTreeRoot root){\n+ return compress(mb, k, new CompressionSettingsBuilder().create());\n+ }\n+\n/**\n* The main method for compressing the input matrix.\n*\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysds/runtime/compress/workload/AWTreeNode.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysds.runtime.compress.workload;\n+\n+import java.util.ArrayList;\n+import java.util.List;\n+\n+/**\n+ * A workload tree is a compact representation of the operations on a matrix and derived intermediates, including the\n+ * basic control structure and inlined functions as well as links to categories.\n+ *\n+ * The intension is to provide the ability to look at a variable and the methods performed on this variable, pruning\n+ * away the rest of the DAG.\n+ *\n+ */\n+public abstract class AWTreeNode {\n+\n+ public enum WTNodeType {\n+ ROOT, FCALL, IF, WHILE, FOR, PARFOR, BASIC_BLOCK;\n+\n+ public boolean isLoop() {\n+ return this == WHILE || this == FOR || this == PARFOR;\n+ }\n+ }\n+\n+ private final WTNodeType _type;\n+ private final List<WTreeNode> _children = new ArrayList<>();\n+\n+ public AWTreeNode(WTNodeType type) {\n+ _type = type;\n+ }\n+\n+ public WTNodeType getType() {\n+ return _type;\n+ }\n+\n+ public List<WTreeNode> getChildNodes() {\n+ return _children;\n+ }\n+\n+ public void addChild(WTreeNode node) {\n+ _children.add(node);\n+ }\n+\n+ public boolean isEmpty(){\n+ return _children.isEmpty();\n+ }\n+\n+ protected String explain(int level) {\n+ StringBuilder sb = new StringBuilder();\n+ // append indentation\n+ for(int i = 0; i < level; i++)\n+ sb.append(\"--\");\n+ // append node summary\n+ sb.append(_type.name());\n+ sb.append(\"\\n\");\n+ // append child nodes\n+ if(!_children.isEmpty())\n+ for(AWTreeNode n : _children)\n+ sb.append(n.explain(level + 1));\n+ return sb.toString();\n+ }\n+\n+ @Override\n+ public String toString() {\n+ StringBuilder sb = new StringBuilder(\"Workload Tree:\\n\");\n+ sb.append(\"--------------------------------------------------------------------------------\\n\");\n+ sb.append(this.explain(1));\n+ sb.append(\"--------------------------------------------------------------------------------\\n\");\n+ return sb.toString();\n+ }\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/workload/WTreeNode.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/workload/WTreeNode.java", "diff": "@@ -25,100 +25,39 @@ import java.util.List;\nimport org.apache.sysds.hops.Hop;\n/**\n- * A workload tree is a compact representation of the operations\n- * on a compressed matrix and derived intermediates, including\n- * the basic control structure and inlined functions as well\n- * as links to categories\n- *\n- * TODO separate classes for inner and leaf nodes?\n+ * A Node in the WTree, this is used for any nodes that are not the root.\n*/\n-public class WTreeNode\n-{\n- public enum WTNodeType{\n- MAIN,\n- FCALL,\n- IF,\n- WHILE,\n- FOR,\n- PARFOR,\n- BASIC_BLOCK;\n- public boolean isLoop() {\n- return this == WHILE ||\n- this == FOR || this == PARFOR;\n- }\n- }\n+public class WTreeNode extends AWTreeNode {\n- private final WTNodeType _type;\n- private final List<WTreeNode> _childs = new ArrayList<>();\n- private final List<Hop> _cops = new ArrayList<>();\n- private int _beginLine = -1;\n- private int _endLine = -1;\n+ private final List<Hop> _ops = new ArrayList<>();\npublic WTreeNode(WTNodeType type) {\n- _type = type;\n- }\n-\n- public WTNodeType getType() {\n- return _type;\n+ super(type);\n}\n- public List<WTreeNode> getChildNodes() {\n- return _childs;\n+ public List<Hop> getOps() {\n+ return _ops;\n}\n- public void addChild(WTreeNode node) {\n- _childs.add(node);\n+ public void addOp(Hop hop) {\n+ _ops.add(hop);\n}\n- public List<Hop> getCompressedOps() {\n- return _cops;\n- }\n-\n- public void addCompressedOp(Hop hop) {\n- _cops.add(hop);\n- }\n-\n- public void setLineNumbers(int begin, int end) {\n- _beginLine = begin;\n- _endLine = end;\n+ @Override\n+ public boolean isEmpty(){\n+ return _ops.isEmpty() && super.isEmpty();\n}\n- public String explain(int level) {\n+ @Override\n+ protected String explain(int level){\nStringBuilder sb = new StringBuilder();\n- //append indentation\n- for( int i=0; i<level; i++ )\n- sb.append(\"--\");\n- //append node summary\n- sb.append(_type.name());\n- if( _beginLine>=0 && _endLine>=0 ) {\n- sb.append(\" (lines \");\n- sb.append(_beginLine);\n- sb.append(\"-\");\n- sb.append(_endLine);\n- sb.append(\")\");\n- }\n- sb.append(\"\\n\");\n- //append child nodes\n- if( !_childs.isEmpty() )\n- for( WTreeNode n : _childs )\n- sb.append(n.explain(level+1));\n- else if( !_cops.isEmpty() ) {\n- for( Hop hop : _cops ) {\n+ sb.append(super.explain(level));\n+ for(Hop hop : _ops) {\nfor(int i = 0; i < level + 1; i++)\nsb.append(\"--\");\nsb.append(hop.toString());\nsb.append(\"\\n\");\n}\n- }\n- return sb.toString();\n- }\n-\n- @Override\n- public String toString() {\n- StringBuilder sb = new StringBuilder(\"Workload Tree:\\n\");\n- sb.append(\"--------------------------------------------------------------------------------\\n\");\n- sb.append(this.explain(1));\n- sb.append(\"--------------------------------------------------------------------------------\\n\");\nreturn sb.toString();\n}\n}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysds/runtime/compress/workload/WTreeRoot.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysds.runtime.compress.workload;\n+\n+import org.apache.sysds.hops.Hop;\n+\n+/**\n+ * The root node of the tree, located at the top of the tree.\n+ *\n+ * This represent a single Hop that have a result that is used on subsequent operations.\n+ */\n+public class WTreeRoot extends AWTreeNode {\n+\n+ private final Hop _root;\n+\n+ public WTreeRoot(Hop root) {\n+ super(WTNodeType.ROOT);\n+ _root = root;\n+ }\n+\n+ /**\n+ * Get the Root hop instruction, that is producing a result used in the rest of the tree.\n+ *\n+ * @return The root hop\n+ */\n+ public Hop getRoot() {\n+ return _root;\n+ }\n+\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/workload/WorkloadAnalyzer.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/workload/WorkloadAnalyzer.java", "diff": "@@ -44,19 +44,19 @@ import org.apache.sysds.parser.ParForStatementBlock;\nimport org.apache.sysds.parser.StatementBlock;\nimport org.apache.sysds.parser.WhileStatement;\nimport org.apache.sysds.parser.WhileStatementBlock;\n-import org.apache.sysds.runtime.compress.workload.WTreeNode.WTNodeType;\n+import org.apache.sysds.runtime.compress.workload.AWTreeNode.WTNodeType;\npublic class WorkloadAnalyzer {\n- public static Map<Long, WTreeNode> getAllCandidateWorkloads(DMLProgram prog) {\n+ public static Map<Long, WTreeRoot> getAllCandidateWorkloads(DMLProgram prog) {\n// extract all compression candidates from program\nList<Hop> candidates = getCandidates(prog);\n// for each candidate, create pruned workload tree\n// TODO memoization of processed subtree if overlap\n- Map<Long, WTreeNode> map = new HashMap<>();\n+ Map<Long, WTreeRoot> map = new HashMap<>();\nfor(Hop cand : candidates) {\n- WTreeNode tree = createWorkloadTree(prog, cand);\n+ WTreeRoot tree = createWorkloadTree(prog, cand);\npruneWorkloadTree(tree);\nmap.put(cand.getHopID(), tree);\n}\n@@ -64,15 +64,15 @@ public class WorkloadAnalyzer {\nreturn map;\n}\n- public static List<Hop> getCandidates(DMLProgram prog) {\n+ private static List<Hop> getCandidates(DMLProgram prog) {\nList<Hop> candidates = new ArrayList<>();\nfor(StatementBlock sb : prog.getStatementBlocks())\ngetCandidates(sb, prog, candidates, new HashSet<>());\nreturn candidates;\n}\n- public static WTreeNode createWorkloadTree(DMLProgram prog, Hop candidate) {\n- WTreeNode main = new WTreeNode(WTNodeType.MAIN);\n+ private static WTreeRoot createWorkloadTree(DMLProgram prog, Hop candidate) {\n+ WTreeRoot main = new WTreeRoot(candidate);\n// TODO generalize, below line assumes only pread candidates (at bottom on DAGs)\nSet<String> compressed = new HashSet<>();\ncompressed.add(candidate.getName());\n@@ -81,7 +81,7 @@ public class WorkloadAnalyzer {\nreturn main;\n}\n- public static boolean pruneWorkloadTree(WTreeNode node) {\n+ private static boolean pruneWorkloadTree(AWTreeNode node) {\n// recursively process sub trees\nIterator<WTreeNode> iter = node.getChildNodes().iterator();\nwhile(iter.hasNext()) {\n@@ -90,8 +90,7 @@ public class WorkloadAnalyzer {\n}\n// indicate that node can be removed\n- return node.getChildNodes().isEmpty()\n- && node.getCompressedOps().isEmpty();\n+ return node.isEmpty();\n}\nprivate static void getCandidates(StatementBlock sb, DMLProgram prog, List<Hop> cands, Set<String> fStack) {\n@@ -157,7 +156,8 @@ public class WorkloadAnalyzer {\nhop.setVisited();\n}\n- private static WTreeNode createWorkloadTree(StatementBlock sb, DMLProgram prog, Set<String> compressed, Set<String> fStack) {\n+ private static WTreeNode createWorkloadTree(StatementBlock sb, DMLProgram prog, Set<String> compressed,\n+ Set<String> fStack) {\nWTreeNode node = null;\nif(sb instanceof FunctionStatementBlock) {\nFunctionStatementBlock fsb = (FunctionStatementBlock) sb;\n@@ -231,11 +231,12 @@ public class WorkloadAnalyzer {\nHop.resetVisitStatus(sb.getHops());\n}\n}\n- node.setLineNumbers(sb.getBeginLine(), sb.getEndLine());\n+\nreturn node;\n}\n- private static void createWorkloadTree(Hop hop, DMLProgram prog, WTreeNode parent, Set<String> compressed, Set<String> fStack) {\n+ private static void createWorkloadTree(Hop hop, DMLProgram prog, WTreeNode parent, Set<String> compressed,\n+ Set<String> fStack) {\nif(hop == null)\nreturn;\nhop.resetVisitStatus();\n@@ -243,7 +244,8 @@ public class WorkloadAnalyzer {\nhop.resetVisitStatus();\n}\n- private static void createWorkloadTree(Hop hop, DMLProgram prog, WTreeNode parent, Set<String> compressed, Set<Long> compressed2, Set<String> fStack) {\n+ private static void createWorkloadTree(Hop hop, DMLProgram prog, WTreeNode parent, Set<String> compressed,\n+ Set<Long> compressed2, Set<String> fStack) {\nif(hop == null || hop.isVisited())\nreturn;\n@@ -252,8 +254,8 @@ public class WorkloadAnalyzer {\ncreateWorkloadTree(c, prog, parent, compressed, compressed2, fStack);\n// map statement block propagation to hop propagation\n- if( HopRewriteUtils.isData(hop, OpOpData.PERSISTENTREAD, OpOpData.TRANSIENTREAD)\n- && compressed.contains(hop.getName()) ) {\n+ if(HopRewriteUtils.isData(hop, OpOpData.PERSISTENTREAD, OpOpData.TRANSIENTREAD) &&\n+ compressed.contains(hop.getName())) {\ncompressed2.add(hop.getHopID());\n}\n@@ -261,15 +263,12 @@ public class WorkloadAnalyzer {\n// if any input is compressed we collect this hop as a compressed operation\nif(hop.getInput().stream().anyMatch(h -> compressed2.contains(h.getHopID()))) {\nif(!HopRewriteUtils.isData(hop, OpOpData.PERSISTENTREAD, // all, but data ops\n- OpOpData.TRANSIENTREAD, OpOpData.TRANSIENTWRITE) )\n- {\n- parent.addCompressedOp(hop);\n+ OpOpData.TRANSIENTREAD, OpOpData.TRANSIENTWRITE)) {\n+ parent.addOp(hop);\n}\n// if the output size also qualifies for compression, we propagate this status\n- if( RewriteCompressedReblock.satisfiesSizeConstraintsForCompression(hop)\n- && hop.getDataType().isMatrix() )\n- {\n+ if(RewriteCompressedReblock.satisfiesSizeConstraintsForCompression(hop) && hop.getDataType().isMatrix()) {\ncompressed2.add(hop.getHopID());\n}\n}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/SingletonLookupHashMap.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysds.runtime.controlprogram;\n+\n+import java.util.HashMap;\n+\n+/**\n+ * This class allows sharing of objects across the entire program.\n+ *\n+ * It is used for instance for sharing WTrees for workload aware compression\n+ */\n+public final class SingletonLookupHashMap {\n+ // Shared singleton map\n+ private static SingletonLookupHashMap singleton = new SingletonLookupHashMap();\n+\n+ private final HashMap<Integer, Object> map;\n+\n+ private SingletonLookupHashMap() {\n+ map = new HashMap<>();\n+ }\n+\n+ public Object get(int id) {\n+ return map.get(id);\n+ }\n+\n+ public int put(Object obj) {\n+ int key = obj.hashCode();\n+ while(map.containsKey(key))\n+ key++; // linear try again until empty key is found.\n+\n+ map.put(key, obj);\n+ return key;\n+ }\n+\n+ public boolean containsKey(int id) {\n+ return map.containsKey(id);\n+ }\n+\n+ public void removeKey(int id) {\n+ map.remove(id);\n+ }\n+\n+ @Override\n+ public String toString() {\n+ return map.toString();\n+ }\n+\n+ public static final SingletonLookupHashMap getMap() {\n+ return singleton;\n+ }\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/CompressionCPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/CompressionCPInstruction.java", "diff": "@@ -21,6 +21,8 @@ package org.apache.sysds.runtime.instructions.cp;\nimport org.apache.sysds.hops.OptimizerUtils;\nimport org.apache.sysds.runtime.compress.CompressedMatrixBlockFactory;\n+import org.apache.sysds.runtime.compress.workload.WTreeRoot;\n+import org.apache.sysds.runtime.controlprogram.SingletonLookupHashMap;\nimport org.apache.sysds.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysds.runtime.instructions.InstructionUtils;\nimport org.apache.sysds.runtime.matrix.data.MatrixBlock;\n@@ -28,8 +30,12 @@ import org.apache.sysds.runtime.matrix.operators.Operator;\npublic class CompressionCPInstruction extends ComputationCPInstruction {\n- private CompressionCPInstruction(Operator op, CPOperand in, CPOperand out, String opcode, String istr) {\n+ private final int _singletonLookupID;\n+\n+ private CompressionCPInstruction(Operator op, CPOperand in, CPOperand out, String opcode, String istr,\n+ int singletonLookupID) {\nsuper(CPType.Compression, op, in, null, null, out, opcode, istr);\n+ this._singletonLookupID = singletonLookupID;\n}\npublic static CompressionCPInstruction parseInstruction(String str) {\n@@ -37,15 +43,27 @@ public class CompressionCPInstruction extends ComputationCPInstruction {\nString opcode = parts[0];\nCPOperand in1 = new CPOperand(parts[1]);\nCPOperand out = new CPOperand(parts[2]);\n- return new CompressionCPInstruction(null, in1, out, opcode, str);\n+ if(parts.length == 4) {\n+ int treeNodeID = Integer.parseInt(parts[3]);\n+ return new CompressionCPInstruction(null, in1, out, opcode, str, treeNodeID);\n+ }\n+ else {\n+ return new CompressionCPInstruction(null, in1, out, opcode, str, 0);\n+ }\n}\n@Override\npublic void processInstruction(ExecutionContext ec) {\n// Get matrix block input\nMatrixBlock in = ec.getMatrixInput(input1.getName());\n+ SingletonLookupHashMap m = SingletonLookupHashMap.getMap();\n+\n+ WTreeRoot root = (_singletonLookupID != 0) ? (WTreeRoot) m.get(_singletonLookupID) : null;\n// Compress the matrix block\n- MatrixBlock out = CompressedMatrixBlockFactory.compress(in, OptimizerUtils.getConstrainedNumThreads(-1)).getLeft();\n+ MatrixBlock out = CompressedMatrixBlockFactory.compress(in, OptimizerUtils.getConstrainedNumThreads(-1), root)\n+ .getLeft();\n+\n+ m.removeKey(_singletonLookupID);\n// Set output and release input\nec.releaseMatrixInput(input1.getName());\nec.setMatrixOutput(output.getName(), out);\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/compress/WorkloadAnalysisTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/compress/WorkloadAnalysisTest.java", "diff": "@@ -26,11 +26,14 @@ import org.apache.sysds.hops.ipa.InterProceduralAnalysis;\nimport org.apache.sysds.test.AutomatedTestBase;\nimport org.apache.sysds.test.TestConfiguration;\nimport org.apache.sysds.test.TestUtils;\n+import org.apache.sysds.utils.Statistics;\nimport org.junit.Assert;\nimport org.junit.Test;\n-public class WorkloadAnalysisTest extends AutomatedTestBase\n-{\n+public class WorkloadAnalysisTest extends AutomatedTestBase {\n+\n+ // private static final Log LOG = LogFactory.getLog(WorkloadAnalysisTest.class.getName());\n+\nprivate final static String TEST_NAME1 = \"WorkloadAnalysisMlogreg\";\nprivate final static String TEST_NAME2 = \"WorkloadAnalysisLm\";\nprivate final static String TEST_DIR = \"functions/compress/\";\n@@ -45,27 +48,25 @@ public class WorkloadAnalysisTest extends AutomatedTestBase\n@Test\npublic void testMlogregCP() {\n- runWorkloadAnalysisTest(TEST_NAME1, ExecMode.HYBRID);\n+ runWorkloadAnalysisTest(TEST_NAME1, ExecMode.HYBRID, 2);\n}\n@Test\npublic void testLmCP() {\n- runWorkloadAnalysisTest(TEST_NAME2, ExecMode.HYBRID);\n+ runWorkloadAnalysisTest(TEST_NAME2, ExecMode.HYBRID, 2);\n}\n- private void runWorkloadAnalysisTest(String testname, ExecMode mode)\n- {\n+ private void runWorkloadAnalysisTest(String testname, ExecMode mode, int compressionCount) {\nExecMode oldPlatform = setExecMode(mode);\nboolean oldFlag = InterProceduralAnalysis.CLA_WORKLOAD_ANALYSIS;\n- try\n- {\n+ try {\nloadTestConfiguration(getTestConfiguration(testname));\nInterProceduralAnalysis.CLA_WORKLOAD_ANALYSIS = true;\nString HOME = SCRIPT_DIR + TEST_DIR;\nfullDMLScriptName = HOME + testname + \".dml\";\n- programArgs = new String[]{\"-stats\",\"-args\", input(\"X\"), input(\"y\"), output(\"B\") };\n+ programArgs = new String[] {\"-stats\", \"40\", \"-args\", input(\"X\"), input(\"y\"), output(\"B\")};\ndouble[][] X = getRandomMatrix(10000, 20, 0, 1, 1.0, 7);\nwriteInputMatrixWithMTD(\"X\", X, false);\n@@ -73,11 +74,13 @@ public class WorkloadAnalysisTest extends AutomatedTestBase\nwriteInputMatrixWithMTD(\"y\", y, false);\nrunTest(true, false, null, -1);\n- //TODO check for compressed operations\n- //(right now test only checks that the workload analysis does not crash)\n// check various additional expectations\n+ long actualCompressionCount = Statistics.getCPHeavyHitterCount(\"compress\");\n+ Assert.assertEquals(compressionCount, actualCompressionCount);\n+ Assert.assertTrue(heavyHittersContainsString(\"compress\"));\nAssert.assertFalse(heavyHittersContainsString(\"m_scale\"));\n+\n}\nfinally {\nresetExecMode(oldPlatform);\n@@ -87,6 +90,6 @@ public class WorkloadAnalysisTest extends AutomatedTestBase\n@Override\nprotected File getConfigTemplateFile() {\n- return new File(SCRIPT_DIR + TEST_DIR + \"force\", \"SystemDS-config-compress.xml\");\n+ return new File(SCRIPT_DIR + TEST_DIR + \"force\", \"SystemDS-config-compress-workload.xml\");\n}\n}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/compress/force/SystemDS-config-compress-workload.xml", "diff": "+<!--\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+-->\n+\n+<root>\n+ <sysds.compressed.linalg>workload</sysds.compressed.linalg>\n+ <sysds.cp.parallel.ops>true</sysds.cp.parallel.ops>\n+ <sysds.scratch>target/force_comp_scratch_space</sysds.scratch>\n+</root>\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/CompressedMatrixBlock.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/CompressedMatrixBlock.java", "diff": "@@ -24,6 +24,7 @@ import java.io.DataOutput;\nimport java.io.IOException;\nimport java.io.ObjectInput;\nimport java.io.ObjectOutput;\n+import java.lang.ref.SoftReference;\nimport java.util.ArrayList;\nimport java.util.Arrays;\nimport java.util.List;\n@@ -119,6 +120,11 @@ public class CompressedMatrixBlock extends MatrixBlock {\n*/\nprotected boolean overlappingColGroups = false;\n+ /**\n+ * Soft reference to a decompressed version of this matrix block.\n+ */\n+ protected SoftReference<MatrixBlock> decompressedVersion;\n+\n/**\n* Constructor for building an empty Compressed Matrix block object.\n*\n@@ -195,6 +201,16 @@ public class CompressedMatrixBlock extends MatrixBlock {\nTiming time = new Timing(true);\n+ if(decompressedVersion != null && decompressedVersion.get() != null){\n+ if(DMLScript.STATISTICS || LOG.isDebugEnabled()) {\n+ double t = time.stop();\n+ LOG.debug(\"decompressed block was in soft reference.\");\n+ DMLCompressionStatistics.addDecompressTime(t, 1);\n+ }\n+ return decompressedVersion.get();\n+ }\n+\n+\nlong nnz = getNonZeros() == -1 ? recomputeNonZeros() : nonZeros;\nif(isEmpty())\nreturn new MatrixBlock(rlen, clen, true, 0);\n@@ -215,6 +231,8 @@ public class CompressedMatrixBlock extends MatrixBlock {\nLOG.debug(\"decompressed block w/ k=\" + 1 + \" in \" + t + \"ms.\");\nDMLCompressionStatistics.addDecompressTime(t, 1);\n}\n+\n+ decompressedVersion = new SoftReference<>(ret);\nreturn ret;\n}\n@@ -245,6 +263,16 @@ public class CompressedMatrixBlock extends MatrixBlock {\nreturn decompress();\nTiming time = new Timing(true);\n+\n+ if(decompressedVersion != null && decompressedVersion.get() != null){\n+ if(DMLScript.STATISTICS || LOG.isDebugEnabled()) {\n+ double t = time.stop();\n+ LOG.debug(\"decompressed block was in soft reference.\");\n+ DMLCompressionStatistics.addDecompressTime(t, k);\n+ }\n+ return decompressedVersion.get();\n+ }\n+\nMatrixBlock ret = getUncompressedColGroupAndRemoveFromListOfColGroups();\nif(ret != null && getColGroups().size() == 0)\nreturn ret;\n@@ -260,6 +288,7 @@ public class CompressedMatrixBlock extends MatrixBlock {\nDMLCompressionStatistics.addDecompressTime(t, k);\n}\n+ decompressedVersion = new SoftReference<>(ret);\nreturn ret;\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/lib/CLALibBinaryCellOp.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/lib/CLALibBinaryCellOp.java", "diff": "@@ -92,17 +92,7 @@ public class CLALibBinaryCellOp {\nMatrixBlock d_compressed = m1.decompress(op.getNumThreads());\nLibMatrixBincell.bincellOpInPlace(d_compressed, that, op);\nreturn d_compressed;\n- // if(left) {\n- // return that.binaryOperations(op, d_compressed, result);\n- // }\n- // else {\n- // return d_compressed.binaryOperations(op, that, result);\n- // }\n- }\n- // else if(that.isInSparseFormat())\n- // return binaryMMSparse(m1, that, op, left);\n- // else\n- // return binaryMMDense(m1, that, op, left);\n+ }\n}\nelse if(isSupportedBinaryCellOp(op.fn))\nreturn bincellOp(m1, that, setupCompressedReturnMatrixBlock(m1, result), op, left);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupValue.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupValue.java", "diff": "@@ -22,6 +22,7 @@ package org.apache.sysds.runtime.compress.colgroup;\nimport java.io.DataInput;\nimport java.io.DataOutput;\nimport java.io.IOException;\n+import java.lang.ref.SoftReference;\nimport java.util.Arrays;\nimport java.util.HashSet;\nimport java.util.Set;\n@@ -78,7 +79,7 @@ public abstract class ColGroupValue extends ColGroupCompressed implements Clonea\nprotected ADictionary _dict;\n/** The count of each distinct value contained in the dictionary */\n- private int[] counts;\n+ private SoftReference<int[]> counts;\nprotected ColGroupValue(int numRows) {\nsuper(numRows);\n@@ -92,7 +93,7 @@ public abstract class ColGroupValue extends ColGroupCompressed implements Clonea\nprotected ColGroupValue(int[] colIndices, int numRows, ADictionary dict, int[] cachedCounts) {\nsuper(colIndices, numRows);\n_dict = dict;\n- counts = cachedCounts;\n+ counts = new SoftReference<>(cachedCounts);\n}\n@Override\n@@ -183,18 +184,29 @@ public abstract class ColGroupValue extends ColGroupCompressed implements Clonea\n* @return the count of each value in the MatrixBlock.\n*/\npublic final int[] getCounts() {\n-\n- if(counts == null && _dict != null) {\n- counts = getCounts(new int[getNumValues() + (_zeros ? 1 : 0)]);\n- return counts;\n+ int[] countsActual = null;\n+ if(_dict != null) {\n+ if(counts == null || counts.get() == null) {\n+ countsActual = getCounts(new int[getNumValues() + (_zeros ? 1 : 0)]);\n+ counts = new SoftReference<>(countsActual);\n}\nelse\n- return counts;\n+ countsActual = counts.get();\n+\n+ }\n+\n+ return countsActual;\n}\n+ /**\n+ * Get the cached counts. If they are not materialized or the garbage collector have removed them, then null is\n+ * returned\n+ *\n+ * @return the counts or null.\n+ */\npublic final int[] getCachedCounts() {\n- return counts;\n+ return counts != null ? counts.get() : null;\n}\n/**\n@@ -1059,7 +1071,6 @@ public abstract class ColGroupValue extends ColGroupCompressed implements Clonea\npublic final MatrixBlock leftMultByPreAggregateMatrix(MatrixBlock preAgg) {\n-\n// Allocate temporary matrix to multiply into.\nfinal int tmpCol = _colIndexes.length;\nfinal int tmpRow = preAgg.getNumRows();\n@@ -1077,7 +1088,6 @@ public abstract class ColGroupValue extends ColGroupCompressed implements Clonea\nreturn leftMultByPreAggregateMatrix(preAgg, tmpRes);\n}\n-\npublic final MatrixBlock leftMultByPreAggregateMatrix(MatrixBlock preAgg, MatrixBlock tmpRes) {\n// Get dictionary.\nMatrixBlock dictM = forceMatrixBlockDictionary().getMatrixBlock();\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/matrix/data/FrameBlock.java", "new_path": "src/main/java/org/apache/sysds/runtime/matrix/data/FrameBlock.java", "diff": "@@ -2212,6 +2212,7 @@ public class FrameBlock implements CacheBlock, Externalizable {\n// construct class code\nsb.append(\"import org.apache.sysds.runtime.util.UtilFunctions;\\n\");\n+ sb.append(\"import org.apache.sysds.runtime.util.PorterStemmer;\\n\");\nsb.append(\"import org.apache.sysds.runtime.matrix.data.FrameBlock.FrameMapFunction;\\n\");\nsb.append(\"public class \" + cname + \" extends FrameMapFunction {\\n\");\nif(varname.length == 1) {\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysds/runtime/util/PorterStemmer.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+package org.apache.sysds.runtime.util;\n+\n+import org.apache.commons.lang3.StringUtils;\n+\n+import java.util.HashMap;\n+import java.util.Iterator;\n+import java.util.Map;\n+\n+/**\n+ * Stemmer, implementing the Porter Stemming Algorithm\n+ *\n+ * The Stemmer class transforms a word into its root form. The input\n+ * word can be provided a character at time (by calling add()), or at once\n+ * by calling one of the various stem(something) methods.\n+ */\n+\n+public class PorterStemmer\n+{\n+ /* m() measures the number of consonant sequences between 0 and j. if c is\n+ a consonant sequence and v a vowel sequence, and <..> indicates arbitrary\n+ presence,\n+\n+ <c><v> gives 0\n+ <c>vc<v> gives 1\n+ <c>vcvc<v> gives 2\n+ <c>vcvcvc<v> gives 3\n+ ....\n+ */\n+\n+ private static int calcM(String word)\n+ {\n+ int l = word.length() ;\n+ int count = 0;\n+ boolean currentConst = false;\n+ for(int c = 0; c < l; c++) {\n+ if(cons(word, c))\n+ {\n+ if(!currentConst && c != 0) {\n+ count += 1;\n+ }\n+ currentConst = true;\n+ }\n+ else\n+ currentConst = false;\n+\n+ }\n+ return count;\n+ }\n+\n+ /* doublec(j) is true <=> j,(j-1) contain a double consonant. */\n+\n+ private static boolean doublec(String word)\n+ { int len = word.length() - 1;\n+ if (len < 1) return false;\n+ if (word.charAt(len) != word.charAt(len - 1)) return false;\n+ return cons(word, len);\n+ }\n+\n+ /* cvc(i) is true <=> i-2,i-1,i has the form consonant - vowel - consonant\n+ and also if the second c is not w,x or y. this is used when trying to\n+ restore an e at the end of a short word. e.g.\n+\n+ cav(e), lov(e), hop(e), crim(e), but\n+ snow, box, tray.\n+* */\n+ private static boolean cvc(String word)\n+ {\n+ int len = word.length();\n+ int l = len - 1;\n+ if (len < 3)\n+ return false;\n+ if(!cons(word, l) | cons(word, l-1) | !cons(word, (l-2)))\n+ return false;\n+ String ch = String.valueOf(word.charAt(l));\n+ String exceptions = \"wxy\";\n+ return !exceptions.contains(ch);\n+ }\n+\n+ /* vowelinstem() is true <=> 0,...j contains a vowel */\n+ private static boolean vowelinStem(String word, String suffix) {\n+ int length = word.length() - suffix.length();\n+ for(int i=0; i<length; i++)\n+ if(!cons(word, i))\n+ return true;\n+\n+ return false;\n+ }\n+\n+ /* cons(i) is true <=> b[i] is a consonant. */\n+\n+ private static boolean cons(String stem, int i)\n+ {\n+ String vowels = \"aeiou\";\n+ char ch = stem.charAt(i);\n+ if(vowels.contains(String.valueOf(stem.charAt(i))))\n+ return false;\n+ if(ch == 'y')\n+ {\n+ if(i == 0)\n+ return true;\n+ else\n+ return (!cons(stem, i - 1));\n+ }\n+ return true;\n+ }\n+ // process the collection of tuples to find which prefix matches the case.\n+ private static String processMatched(String word, HashMap suffixAndfix, int mCount)\n+ {\n+ String stemmed = null;\n+ Iterator it = suffixAndfix.entrySet().iterator();\n+ while (it.hasNext() && (stemmed == null)) {\n+ Map.Entry pair = (Map.Entry)it.next();\n+ stemmed = replacer(word, pair.getKey().toString(), pair.getValue().toString(), mCount);\n+ it.remove();\n+ }\n+ return stemmed;\n+ }\n+\n+ // replace the suffix with suggeston\n+ private static String replacer(String word, String orig, String replace, int mCount)\n+ {\n+ int l = word.length();\n+ int suffixLength = orig.length();\n+\n+ if (word.endsWith(orig))\n+ {\n+ String stem = word.substring( 0, l - suffixLength);\n+ int m = calcM( stem );\n+ if (m > mCount)\n+ return stem.concat(replace);\n+ else\n+ return word;\n+\n+ }\n+\n+ return null;\n+ }\n+\n+ /* step1() gets rid of plurals and -ed or -ing. e.g.\n+ i.e., condition & suffix -> replacement\n+ SSES -> SS\n+ IES -> I\n+ SS -> SS\n+ S -> \"\"\n+ (m > 0) EED -> EE\n+ vowelSequence(ED) -> \"\"\n+ vowelsequence(ING) -> \"\"\n+ any(\"at, bl, iz\") -> add(e)\n+ doubleconsonant and not(\"l\", \"s\", \"z\") -> remove single letter from end\n+ (m == 1 and cvc) -> add(e)\n+ turns terminal y to i when there is another vowel in the stem.\n+ */\n+\n+ private static String step1(String word)\n+ {\n+ boolean flag = false;\n+ if (word.endsWith(\"s\"))\n+ {\n+ if (word.endsWith(\"sses\"))\n+ word = StringUtils.removeEnd(word, \"es\");\n+ else if (word.endsWith(\"ies\")) {\n+ word = StringUtils.removeEnd(word, \"ies\").concat(\"i\");\n+ }\n+ else if (!word.endsWith(\"ss\") && word.endsWith(\"s\"))\n+ word = StringUtils.removeEnd(word, \"s\");\n+ }\n+ if (word.endsWith(\"eed\"))\n+ {\n+ if (calcM(word) > 1)\n+ word = StringUtils.removeEnd(word, \"d\");\n+ }\n+ else if(word.endsWith(\"ed\") && vowelinStem(word, \"ed\")) {\n+ word = StringUtils.removeEnd(word, \"ed\");\n+ flag = true;\n+ }\n+ else if(word.endsWith(\"ing\") && vowelinStem(word, \"ing\"))\n+ {\n+ word = StringUtils.removeEnd(word, \"ing\");\n+\n+ flag = true;\n+ }\n+\n+ if (flag)\n+ {\n+ if(word.endsWith(\"at\") || word.endsWith(\"bl\") || word.endsWith(\"iz\"))\n+ word = word.concat(\"e\");\n+ int m = calcM(word);\n+ String last = String.valueOf(word.charAt(word.length() - 1));\n+ if (doublec(word) && !\"lsz\".contains(last))\n+ word = word.substring(0, word.length() - 1);\n+ else if (m == 1 && cvc(word))\n+ word = word.concat(\"e\");\n+ }\n+ if (word.endsWith(\"y\") && vowelinStem(word, \"y\"))\n+ word = StringUtils.removeEnd(word, \"y\").concat(\"i\");\n+\n+ return word;\n+ }\n+\n+ // step2() maps double suffices to single ones\n+\n+ private static String step2(String word) {\n+ int len = word .length();\n+ if (len == 0) return word;\n+ HashMap<String, String> suffixAndfix = new HashMap<String, String>()\n+ {{\n+ put(\"ational\", \"ate\");\n+ put(\"tional\",\"tion\");\n+ put(\"enci\",\"ence\");\n+ put(\"anci\",\"ance\");\n+ put(\"izer\",\"ize\");\n+ put(\"bli\",\"ble\");\n+ put(\"alli\", \"al\");\n+ put(\"entli\",\"ent\");\n+ put(\"eli\",\"e\");\n+ put(\"ousli\",\"ous\");\n+ put(\"ization\",\"ize\");\n+ put(\"ation\",\"ate\");\n+ put(\"ator\",\"ate\");\n+ put(\"alism\",\"al\");\n+ put(\"iveness\", \"ive\");\n+ put(\"fulness\",\"ful\");\n+ put(\"ousness\", \"ous\");\n+ put(\"aliti\", \"al\");\n+ put(\"iviti\",\"ive\");\n+ put(\"biliti\", \"ble\");\n+ put(\"log\", \"logi\");\n+ put(\"icate\", \"ic\");\n+ put(\"ative\",\"\");\n+ put(\"alize\",\"al\");\n+ put(\"iciti\",\"ic\");\n+ put(\"ical\",\"ic\");\n+ }};\n+\n+ String stemmed = processMatched(word, suffixAndfix, 0);\n+ return (stemmed != null)? stemmed: word;\n+ }\n+ // handles -ic-, -full, -ness etc.\n+ private static String step3(String word) {\n+ int len = word .length();\n+ if (len == 0) return word;\n+ HashMap<String, String> suffixAndfix = new HashMap<String, String>()\n+ {{\n+ put(\"icate\", \"ic\");\n+ put(\"ative\",\"\");\n+ put(\"alize\",\"al\");\n+ put(\"iciti\",\"ic\");\n+ put(\"ical\",\"ic\");\n+ put(\"ful\",\"\");\n+ put(\"ness\",\"\");\n+ }};\n+\n+ String stemmed = processMatched(word, suffixAndfix, 0);\n+ return (stemmed != null)? stemmed: word;\n+\n+ }\n+\n+ // takes off -ant, -ence etc., in context <c>vcvc<v>\n+ private static String step4(String word)\n+ {\n+ // first part.\n+ String[] suffix = new String[] {\"al\", \"ance\", \"ence\", \"er\", \"ic\", \"able\", \"ible\", \"ant\",\n+ \"ement\", \"ment\", \"ent\"};\n+ String stemmed = null;\n+ int i = 0;\n+ while(stemmed == null && i < suffix.length)\n+ {\n+ stemmed = replacer(word, suffix[i], \"\", 1);\n+ i++;\n+ }\n+ // exceptions\n+ if(stemmed == null)\n+ {\n+ if(word.length() > 4)\n+ {\n+ char ch = word.charAt(word.length() - 4);\n+ if(ch == 's' || ch == 't')\n+ {\n+ stemmed = replacer(word, \"ion\", \"\", 1);\n+ }\n+ }\n+ }\n+ // exceptions\n+ if (stemmed == null)\n+ {\n+ suffix = new String[] {\"ou\", \"ism\", \"ate\", \"iti\", \"ous\", \"ive\", \"ize\"};\n+ i = 0;\n+ while(stemmed == null && i < suffix.length)\n+ {\n+ stemmed = replacer(word, suffix[i], \"\", 1);\n+ i++;\n+ }\n+ }\n+\n+ return (stemmed != null)? stemmed: word;\n+ }\n+ // handle the last e and l\n+ private static String step5(String word)\n+ {\n+ String stem = StringUtils.removeEnd(word, \"e\");\n+ if(word.endsWith(\"e\") && calcM(word) > 1)\n+ word = stem;\n+ if(word.endsWith(\"e\") && calcM(word) == 1 && !cvc(stem))\n+ word = stem;\n+ if(word.endsWith(\"l\") && doublec(word) && calcM(word) > 1)\n+ word = word.substring(0, word.length() - 1);\n+\n+ return word;\n+ }\n+ public static String stem (String word)\n+ {\n+ if(word.length() >= 3) {\n+ word = step1(word);\n+ word = step2(word);\n+ word = step3(word);\n+ word = step4(word);\n+ if(word.length() > 0)\n+ word = step5(word);\n+ }\n+ return word;\n+ }\n+}\n\\ No newline at end of file\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysds/test/functions/builtin/BuiltinPorterStemmerTest.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysds.test.functions.builtin;\n+\n+import org.apache.sysds.common.Types;\n+import org.apache.sysds.runtime.matrix.data.FrameBlock;\n+import org.apache.sysds.test.AutomatedTestBase;\n+import org.apache.sysds.test.TestConfiguration;\n+import org.apache.sysds.test.TestUtils;\n+import org.junit.AfterClass;\n+import org.junit.Assert;\n+import org.junit.BeforeClass;\n+import org.junit.Test;\n+\n+import java.io.IOException;\n+\n+public class BuiltinPorterStemmerTest extends AutomatedTestBase {\n+\n+ private final static String TEST_NAME = \"porterStemmerTest\";\n+ private final static String TEST_DIR = \"functions/builtin/\";\n+ private static final String TEST_CLASS_DIR = BuiltinPorterStemmerTest.class.getSimpleName() + \"/\";\n+ private static final String INPUT = DATASET_DIR +\"stemming/dictionary.csv\";\n+\n+ @BeforeClass\n+ public static void init() {\n+ TestUtils.clearDirectory(TEST_DATA_DIR + TEST_CLASS_DIR);\n+ }\n+\n+ @AfterClass\n+ public static void cleanUp() {\n+ if (TEST_CACHE_ENABLED) {\n+ TestUtils.clearDirectory(TEST_DATA_DIR + TEST_CLASS_DIR);\n+ }\n+ }\n+\n+ @Override\n+ public void setUp() {\n+ TestUtils.clearAssertionInformation();\n+ addTestConfiguration(TEST_NAME, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME, new String[] {\"D\"}));\n+ if (TEST_CACHE_ENABLED) {\n+ setOutAndExpectedDeletionDisabled(true);\n+ }\n+ }\n+ @Test\n+ public void testStemmerCP() {\n+ runStemmerTest(Types.ExecMode.SINGLE_NODE);\n+ }\n+\n+ @Test\n+ public void testStemmerSpark() {\n+ runStemmerTest(Types.ExecMode.SPARK);\n+ }\n+\n+ private void runStemmerTest(Types.ExecMode et)\n+ {\n+ Types.ExecMode modeOld = setExecMode(et);\n+ try {\n+ loadTestConfiguration(getTestConfiguration(TEST_NAME));\n+ String HOME = SCRIPT_DIR + TEST_DIR;\n+\n+ fullDMLScriptName = HOME + TEST_NAME + \".dml\";\n+ programArgs = new String[] {\"-args\", INPUT, output(\"S\"), output(\"E\")};\n+\n+ runTest(true, EXCEPTION_NOT_EXPECTED, null, -1);\n+ FrameBlock outputFrame = readDMLFrameFromHDFS(\"S\", Types.FileFormat.CSV);\n+ FrameBlock inputFrame = readDMLFrameFromHDFS(\"E\", Types.FileFormat.CSV);\n+ String[] output = (String[])outputFrame.getColumnData(0);\n+ String[] input = (String[])inputFrame.getColumnData(0);\n+ //expected vs stemmer output\n+ int count = 0;\n+ for(int i = 0; i<input.length; i++) {\n+ if(input[i].equals(output[i]))\n+ count++;\n+ }\n+ Assert.assertEquals(110, count, 10);\n+ }\n+ catch(IOException e) {\n+ e.printStackTrace();\n+ }\n+ finally {\n+ resetExecMode(modeOld);\n+ }\n+ }\n+}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/resources/datasets/stemming/dictionary.csv", "diff": "+Words,Stemmed Equivalents\n+caresses,caress\n+ponies,poni\n+ties,ti\n+caress,caress\n+cats,cat\n+feed,feed\n+agreed,agre\n+disabled,disabl\n+matting,mat\n+mating,mate\n+meeting,meet\n+milling,mill\n+messing,mess\n+meetings,meet\n+abbess,abbess\n+abbey,abbei\n+abbeys,abbei\n+abbominable,abbomin\n+abbot,abbot\n+abbots,abbot\n+abbreviated,abbrevi\n+abdication,abdic\n+abduction,abduct\n+abed,ab\n+abel,abel\n+aberga,aberga\n+abergavenny,abergavenni\n+abet,abet\n+abetting,abet\n+abhominable,abhomin\n+abhor,abhor\n+abhorr,abhorr\n+abhorred,abhor\n+abhorrence,abhorr\n+abhorring,abhor\n+abhors,abhor\n+abhorson,abhorson\n+abide,abid\n+abides,abid\n+abilities,abil\n+ability,abil\n+abingdon,abingdon\n+abject,abject\n+abjectly,abjectli\n+abjects,abject\n+abjur,abjur\n+abjure,abjur\n+abler,abler\n+ablest,ablest\n+aboard,aboard\n+accomplishment,accomplish\n+boiling,boil\n+boils,boil\n+booths,booth\n+booties,booti\n+boundary,boundari\n+bounded,bound\n+cuts,cut\n+cytherea,cytherea\n+dawdling,dawdl\n+deception,decept\n+matting,mat\n+management,manag\n+manager,manag\n+managing,manag\n+manakin,manakin\n+manasseh,manasseh\n+manchen,manchen\n+manchester,manchest\n+manchus,manchu\n+mandate,mandat\n+mandragora,mandragora\n+mandrake,mandrak\n+mandrakes,mandrak\n+mane,mane\n+manent,manent\n+manes,mane\n+manet,manet\n+manfully,manfulli\n+mangelwurzel,mangelwurzel\n+mangle,mangl\n+mangled,mangl\n+mangles,mangl\n+mangling,mangl\n+mangnall,mangnal\n+mango,mango\n+mangoes,mango\n+mangy,mangi\n+manhood,manhood\n+manhoods,manhood\n+mania,mania\n+manifest,manifest\n+manifested,manifest\n+wived,wive\n+wonderfully,wonderfulli\n+wondering,wonder\n+wonders,wonder\n+worms,worm\n+wormwood,wormwood\n+wormy,wormi\n+worn,worn\n+worret,worret\n+worried,worri\n+worries,worri\n+worry,worri\n+worrying,worri\n+worser,worser\n+worship,worship\n+worshipful,worship\n\\ No newline at end of file\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/builtin/porterStemmerTest.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+# input frame contain two columns of actual words and their stemmed equivalents\n+words = read($1, data_type = \"frame\", header = TRUE, format = \"csv\")\n+stemmed = map(words[, 1], \"x -> PorterStemmer.stem(x)\")\n+# write the results from the stemmer\n+write(stemmed, $2, format = \"csv\")\n+# write the dictionary equivalents for matching in java file\n+equ = words[, 2]\n+write(equ, $3, format = \"csv\")\n\\ No newline at end of file\n" } ]

[ { "change_type": "ADD", "old_path": null, "new_path": "scripts/builtin/shortestPath.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+#\n+# Computes the minimum distances (shortest-path) between a single\n+# source vertex and every other vertex in the graph.\n+#\n+# Grzegorz Malewicz, Matthew H. Austern, Aart J. C. Bilk,\n+# James C. Dehnert, Ikkan Horn, Naty Leiser and Grzegorz Czajkowski:\n+# Pregel: A System for Large-Scale Graph Processing\n+#\n+#------------------------------------------------------------------------------\n+# NAME TYPE MEANING\n+# G MATRIX adjacency matrix of the labeled graph: Such graph can be directed\n+# (G is symmetric) or undirected (G is not symmetric).\n+# The values of G can be 0/1 (just specifying whether the nodes\n+# are connected or not) or integer values (representing the weight\n+# of the edges or the distances between nodes, 0 if not connected).\n+#\n+# maxi Integer Integer max number of iterations accepted (0 for FALSE, i.e.\n+# max number of iterations not defined)\n+#\n+# sourceNode Integer node index to calculate the shortest paths to all other nodes.\n+#\n+# verbose Boolean flag for verbose debug output\n+#------------------------------------------------------------------------------\n+# C Matrix Output matrix (double) of minimum distances (shortest-path) between\n+# vertices: The value of the ith row and the jth column of the output\n+# matrix is the minimum distance shortest-path from vertex i to vertex j.\n+# When the value of the minimum distance is infinity, the two nodes are\n+# not connected.\n+#------------------------------------------------------------------------------\n+\n+m_shortestPath = function(Matrix[Double] G, Integer maxi = 0, Integer sourceNode, Boolean verbose = FALSE)\n+ return (Matrix[Double] C)\n+{\n+ if(verbose) {\n+ print(\"SHORTEST PATH CALCULATION\");\n+ }\n+\n+ if(min(G) < 0){\n+ stop(\"All values in G must be positive\")\n+ }\n+\n+ if(ncol(G) != nrow(G)){\n+ stop(\"Not correct matrix dimensions\")\n+ }\n+\n+ matrixSize = nrow(G)\n+\n+ G = replace(target=G, pattern=0, replacement=Inf)\n+\n+ # initialize the matrix of minimum distances with \"infinity\" values:\n+ minDistVector = matrix(Inf,rows=matrixSize,cols=1)\n+\n+ # update minimum distance from the sourceNode to itself to 0:\n+ minDistVector[sourceNode,1] = 0\n+\n+ iter = 1\n+ diff = Inf;\n+ while( diff > 0 & (maxi==0 | iter<=maxi) ) {\n+ w = t(colMins(G + minDistVector))\n+ u = min(w, minDistVector);\n+ diff = sum(u != minDistVector)\n+ minDistVector = u; # update assignment\n+ if( verbose ){\n+ print(\"Shortest Path: iter = \"+iter+\", #diff = \"+diff);\n+ }\n+ iter = iter + 1;\n+ }\n+\n+ C=minDistVector\n+ if(verbose) {\n+ print(\"SHORTEST PATH CALCULATION FINISHED, CHECK OUTPUT MATRIX OF MINIMUM DISTANCES:\");\n+ print(toString(C))\n+ }\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/common/Builtins.java", "new_path": "src/main/java/org/apache/sysds/common/Builtins.java", "diff": "@@ -227,6 +227,7 @@ public enum Builtins {\nSEQ(\"seq\", false),\nSHERLOCK(\"sherlock\", true),\nSHERLOCKPREDICT(\"sherlockPredict\", true),\n+ SHORTESTPATH(\"shortestPath\", true),\nSIGMOID(\"sigmoid\", true), // 1 / (1 + exp(-X))\nSIGN(\"sign\", false),\nSIN(\"sin\", false),\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysds/test/functions/builtin/BuiltinShortestPathTest.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysds.test.functions.builtin;\n+\n+import java.util.HashMap;\n+\n+import org.junit.Test;\n+\n+import org.apache.sysds.runtime.matrix.data.MatrixValue;\n+import org.apache.sysds.test.AutomatedTestBase;\n+import org.apache.sysds.test.TestConfiguration;\n+import org.apache.sysds.test.TestUtils;\n+\n+public class BuiltinShortestPathTest extends AutomatedTestBase {\n+ private final static String TEST_NAME = \"shortestPathTest\";\n+ private final static String TEST_DIR = \"functions/builtin/\";\n+ private static final String TEST_CLASS_DIR = TEST_DIR + BuiltinShortestPathTest.class.getSimpleName() + \"/\";\n+ private final static double eps = 1e-10;\n+\n+ @Override\n+ public void setUp() {\n+ TestUtils.clearAssertionInformation();\n+ addTestConfiguration(TEST_NAME,new TestConfiguration(TEST_CLASS_DIR, TEST_NAME,new String[]{\"r\"}));\n+ }\n+\n+ @Test\n+ public void testShortestPathNode1CP() {\n+ runShortestPathNodeTest(1, new double[][] {{0}, {2}, {5}, {5}});\n+ }\n+\n+ @Test\n+ public void testShortestPathNode2CP() {\n+ runShortestPathNodeTest(2, new double[][] {{1}, {0}, {4}, {5}});\n+ }\n+\n+ @Test\n+ public void testShortestPathNode3CP() {\n+ runShortestPathNodeTest(3, new double[][] {{4}, {3}, {0}, {1}});\n+ }\n+\n+\n+\n+ private void runShortestPathNodeTest(int node, double [][] Res) {\n+ loadTestConfiguration(getTestConfiguration(TEST_NAME));\n+\n+ String HOME = SCRIPT_DIR + TEST_DIR;\n+ fullDMLScriptName = HOME + TEST_NAME + \".dml\";\n+ programArgs = new String[]{ \"-args\",\n+ input(\"X\"), String.valueOf(node), output(\"R\")};\n+\n+ double[][] X = {{0, 2, 5, 5 },\n+ {1, 0, 4, 10},\n+ {0, 3, 0, 1 },\n+ {3, 2, 0, 0 }};\n+ writeInputMatrixWithMTD(\"X\", X, true);\n+\n+ runTest(true, false, null, -1);\n+\n+ HashMap<MatrixValue.CellIndex, Double> dmlfile = readDMLMatrixFromOutputDir(\"R\");\n+ double[][] Y = TestUtils.convertHashMapToDoubleArray(dmlfile);\n+ TestUtils.compareMatrices(Res, Y, eps);\n+ }\n+}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/builtin/shortestPathTest.dml", "diff": "+\n+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+X = read($1)\n+C = shortestPath(G=X,sourceNode = $2)\n+\n+write(C, $3)\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/api/DMLOptions.java", "new_path": "src/main/java/org/apache/sysds/api/DMLOptions.java", "diff": "@@ -52,6 +52,8 @@ public class DMLOptions {\npublic boolean clean = false; // Whether to clean up all SystemDS working directories (FS, DFS)\npublic boolean stats = false; // Whether to record and print the statistics\npublic int statsCount = 10; // Default statistics count\n+ public boolean fedStats = false; // Whether to record and print the federated statistics\n+ public int fedStatsCount = 10; // Default federated statistics count\npublic boolean memStats = false; // max memory statistics\npublic Explain.ExplainType explainType = Explain.ExplainType.NONE; // Whether to print the \"Explain\" and if so, what type\npublic ExecMode execMode = OptimizerUtils.getDefaultExecutionMode(); // Execution mode standalone, MR, Spark or a hybrid\n@@ -85,6 +87,8 @@ public class DMLOptions {\n\", clean=\" + clean +\n\", stats=\" + stats +\n\", statsCount=\" + statsCount +\n+ \", fedStats=\" + fedStats +\n+ \", fedStatsCount=\" + fedStatsCount +\n\", memStats=\" + memStats +\n\", explainType=\" + explainType +\n\", execMode=\" + execMode +\n@@ -193,6 +197,17 @@ public class DMLOptions {\n}\n}\n}\n+ dmlOptions.fedStats = line.hasOption(\"fedStats\");\n+ if (dmlOptions.fedStats) {\n+ String fedStatsCount = line.getOptionValue(\"fedStats\");\n+ if(fedStatsCount != null) {\n+ try {\n+ dmlOptions.fedStatsCount = Integer.parseInt(fedStatsCount);\n+ } catch (NumberFormatException e) {\n+ throw new org.apache.commons.cli.ParseException(\"Invalid argument specified for -fedStats option, must be a valid integer\");\n+ }\n+ }\n+ }\ndmlOptions.memStats = line.hasOption(\"mem\");\ndmlOptions.clean = line.hasOption(\"clean\");\n@@ -265,6 +280,9 @@ public class DMLOptions {\nOption statsOpt = OptionBuilder.withArgName(\"count\")\n.withDescription(\"monitors and reports summary execution statistics; heavy hitter <count> is 10 unless overridden; default off\")\n.hasOptionalArg().create(\"stats\");\n+ Option fedStatsOpt = OptionBuilder.withArgName(\"count\")\n+ .withDescription(\"monitors and reports summary execution statistics of federated workers; heavy hitter <count> is 10 unless overridden; default off\")\n+ .hasOptionalArg().create(\"fedStats\");\nOption memOpt = OptionBuilder.withDescription(\"monitors and reports max memory consumption in CP; default off\")\n.create(\"mem\");\nOption explainOpt = OptionBuilder.withArgName(\"level\")\n@@ -299,6 +317,7 @@ public class DMLOptions {\noptions.addOption(configOpt);\noptions.addOption(cleanOpt);\noptions.addOption(statsOpt);\n+ options.addOption(fedStatsOpt);\noptions.addOption(memOpt);\noptions.addOption(explainOpt);\noptions.addOption(execOpt);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/api/DMLScript.java", "new_path": "src/main/java/org/apache/sysds/api/DMLScript.java", "diff": "@@ -87,6 +87,8 @@ public class DMLScript\npublic static boolean JMLC_MEM_STATISTICS = false; // whether to gather memory use stats in JMLC\npublic static int STATISTICS_COUNT = DMLOptions.defaultOptions.statsCount; // statistics maximum heavy hitter count\npublic static int STATISTICS_MAX_WRAP_LEN = 30; // statistics maximum wrap length\n+ public static boolean FED_STATISTICS = DMLOptions.defaultOptions.fedStats; // whether to print federated statistics\n+ public static int FED_STATISTICS_COUNT = DMLOptions.defaultOptions.fedStatsCount; // federated statistics maximum heavy hitter count\npublic static ExplainType EXPLAIN = DMLOptions.defaultOptions.explainType; // explain type\npublic static String DML_FILE_PATH_ANTLR_PARSER = DMLOptions.defaultOptions.filePath; // filename of dml/pydml script\npublic static String FLOATING_POINT_PRECISION = \"double\"; // data type to use internally\n@@ -214,6 +216,8 @@ public class DMLScript\n{\nSTATISTICS = dmlOptions.stats;\nSTATISTICS_COUNT = dmlOptions.statsCount;\n+ FED_STATISTICS = dmlOptions.fedStats;\n+ FED_STATISTICS_COUNT = dmlOptions.fedStatsCount;\nJMLC_MEM_STATISTICS = dmlOptions.memStats;\nUSE_ACCELERATOR = dmlOptions.gpu;\nFORCE_ACCELERATOR = dmlOptions.forceGPU;\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/federated/FederatedStatistics.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysds.runtime.controlprogram.federated;\n+\n+import java.io.Serializable;\n+import java.net.InetSocketAddress;\n+import java.text.DecimalFormat;\n+import java.util.ArrayList;\n+import java.util.Arrays;\n+import java.util.Comparator;\n+import java.util.HashMap;\n+import java.util.HashSet;\n+import java.util.List;\n+import java.util.Map.Entry;\n+import java.util.Set;\n+import java.util.concurrent.Future;\n+import javax.net.ssl.SSLException;\n+\n+import org.apache.commons.lang3.tuple.ImmutablePair;\n+import org.apache.commons.lang3.tuple.Pair;\n+import org.apache.sysds.api.DMLScript;\n+import org.apache.sysds.runtime.controlprogram.caching.CacheStatistics;\n+import org.apache.sysds.runtime.controlprogram.context.ExecutionContext;\n+import org.apache.sysds.runtime.controlprogram.federated.FederatedData;\n+import org.apache.sysds.runtime.controlprogram.federated.FederatedRequest.RequestType;\n+import org.apache.sysds.runtime.controlprogram.federated.FederatedStatistics.FedStatsCollection.CacheStatsCollection;\n+import org.apache.sysds.runtime.controlprogram.federated.FederatedStatistics.FedStatsCollection.GCStatsCollection;\n+import org.apache.sysds.runtime.DMLRuntimeException;\n+import org.apache.sysds.runtime.instructions.cp.Data;\n+import org.apache.sysds.runtime.lineage.LineageItem;\n+import org.apache.sysds.utils.Statistics;\n+\n+public class FederatedStatistics {\n+ private static Set<Pair<String, Integer>> _fedWorkerAddresses = new HashSet<>();\n+\n+ public static void registerFedWorker(String host, int port) {\n+ _fedWorkerAddresses.add(new ImmutablePair<>(host, new Integer(port)));\n+ }\n+\n+ public static String displayFedWorkers() {\n+ StringBuilder sb = new StringBuilder();\n+ sb.append(\"Federated Worker Addresses:\\n\");\n+ for(Pair<String, Integer> fedAddr : _fedWorkerAddresses) {\n+ sb.append(String.format(\" %s:%d\", fedAddr.getLeft(), fedAddr.getRight().intValue()));\n+ sb.append(\"\\n\");\n+ }\n+ return sb.toString();\n+ }\n+\n+ public static String displayFedStatistics(int numHeavyHitters) {\n+ StringBuilder sb = new StringBuilder();\n+ FedStatsCollection fedStats = collectFedStats();\n+ sb.append(\"SystemDS Federated Statistics:\\n\");\n+ sb.append(displayCacheStats(fedStats.cacheStats));\n+ sb.append(String.format(\"Total JIT compile time:\\t\\t%.3f sec.\\n\", fedStats.jitCompileTime));\n+ sb.append(displayGCStats(fedStats.gcStats));\n+ sb.append(displayHeavyHitters(fedStats.heavyHitters, numHeavyHitters));\n+ return sb.toString();\n+ }\n+\n+ public static String displayCacheStats(CacheStatsCollection csc) {\n+ StringBuilder sb = new StringBuilder();\n+ sb.append(String.format(\"Cache hits (Mem/Li/WB/FS/HDFS):\\t%d/%d/%d/%d/%d.\\n\",\n+ csc.memHits, csc.linHits, csc.fsBuffHits, csc.fsHits, csc.hdfsHits));\n+ sb.append(String.format(\"Cache writes (Li/WB/FS/HDFS):\\t%d/%d/%d/%d.\\n\",\n+ csc.linWrites, csc.fsBuffWrites, csc.fsWrites, csc.hdfsWrites));\n+ sb.append(String.format(\"Cache times (ACQr/m, RLS, EXP):\\t%.3f/%.3f/%.3f/%.3f sec.\\n\",\n+ csc.acqRTime, csc.acqMTime, csc.rlsTime, csc.expTime));\n+ return sb.toString();\n+ }\n+\n+ public static String displayGCStats(GCStatsCollection gcsc) {\n+ StringBuilder sb = new StringBuilder();\n+ sb.append(String.format(\"Total JVM GC count:\\t\\t%d.\\n\", gcsc.gcCount));\n+ sb.append(String.format(\"Total JVM GC time:\\t\\t%.3f sec.\\n\", gcsc.gcTime));\n+ return sb.toString();\n+ }\n+\n+ public static String displayHeavyHitters(HashMap<String, Pair<Long, Double>> heavyHitters) {\n+ return displayHeavyHitters(heavyHitters, 10);\n+ }\n+\n+ public static String displayHeavyHitters(HashMap<String, Pair<Long, Double>> heavyHitters, int num) {\n+ StringBuilder sb = new StringBuilder();\n+ @SuppressWarnings(\"unchecked\")\n+ Entry<String, Pair<Long, Double>>[] hhArr = heavyHitters.entrySet().toArray(new Entry[0]);\n+ Arrays.sort(hhArr, new Comparator<Entry<String, Pair<Long, Double>>>() {\n+ public int compare(Entry<String, Pair<Long, Double>> e1, Entry<String, Pair<Long, Double>> e2) {\n+ return e1.getValue().getRight().compareTo(e2.getValue().getRight());\n+ }\n+ });\n+\n+ sb.append(\"Heavy hitter instructions:\\n\");\n+ final String numCol = \"#\";\n+ final String instCol = \"Instruction\";\n+ final String timeSCol = \"Time(s)\";\n+ final String countCol = \"Count\";\n+ int numHittersToDisplay = Math.min(num, hhArr.length);\n+ int maxNumLen = String.valueOf(numHittersToDisplay).length();\n+ int maxInstLen = instCol.length();\n+ int maxTimeSLen = timeSCol.length();\n+ int maxCountLen = countCol.length();\n+ DecimalFormat sFormat = new DecimalFormat(\"#,##0.000\");\n+ for (int counter = 0; counter < numHittersToDisplay; counter++) {\n+ Entry<String, Pair<Long, Double>> hh = hhArr[hhArr.length - 1 - counter];\n+ String instruction = hh.getKey();\n+ maxInstLen = Math.max(maxInstLen, instruction.length());\n+ String timeString = sFormat.format(hh.getValue().getRight());\n+ maxTimeSLen = Math.max(maxTimeSLen, timeString.length());\n+ maxCountLen = Math.max(maxCountLen, String.valueOf(hh.getValue().getLeft()).length());\n+ }\n+ maxInstLen = Math.min(maxInstLen, DMLScript.STATISTICS_MAX_WRAP_LEN);\n+ sb.append(String.format( \" %\" + maxNumLen + \"s %-\" + maxInstLen + \"s %\"\n+ + maxTimeSLen + \"s %\" + maxCountLen + \"s\", numCol, instCol, timeSCol, countCol));\n+ sb.append(\"\\n\");\n+\n+ for (int counter = 0; counter < numHittersToDisplay; counter++) {\n+ String instruction = hhArr[hhArr.length - 1 - counter].getKey();\n+ String [] wrappedInstruction = Statistics.wrap(instruction, maxInstLen);\n+\n+ String timeSString = sFormat.format(hhArr[hhArr.length - 1 - counter].getValue().getRight());\n+\n+ long count = hhArr[hhArr.length - 1 - counter].getValue().getLeft();\n+ int numLines = wrappedInstruction.length;\n+\n+ for(int wrapIter = 0; wrapIter < numLines; wrapIter++) {\n+ String instStr = (wrapIter < wrappedInstruction.length) ? wrappedInstruction[wrapIter] : \"\";\n+ if(wrapIter == 0) {\n+ sb.append(String.format(\n+ \" %\" + maxNumLen + \"d %-\" + maxInstLen + \"s %\" + maxTimeSLen + \"s %\"\n+ + maxCountLen + \"d\", (counter + 1), instStr, timeSString, count));\n+ }\n+ else {\n+ sb.append(String.format(\n+ \" %\" + maxNumLen + \"s %-\" + maxInstLen + \"s %\" + maxTimeSLen + \"s %\"\n+ + maxCountLen + \"s\", \"\", instStr, \"\", \"\"));\n+ }\n+ sb.append(\"\\n\");\n+ }\n+ }\n+\n+ return sb.toString();\n+ }\n+\n+ private static FedStatsCollection collectFedStats() {\n+ Future<FederatedResponse>[] responses = getFederatedResponses();\n+ FedStatsCollection aggFedStats = new FedStatsCollection();\n+ for(Future<FederatedResponse> res : responses) {\n+ try {\n+ Object[] tmp = res.get().getData();\n+ if(tmp[0] instanceof FedStatsCollection)\n+ aggFedStats.aggregate((FedStatsCollection)tmp[0]);\n+ } catch(Exception e) {\n+ throw new DMLRuntimeException(\"Exception of type \" + e.getClass().toString()\n+ + \" thrown while \" + \"getting the federated stats of the federated response: \", e);\n+ }\n+ }\n+ return aggFedStats;\n+ }\n+\n+ private static Future<FederatedResponse>[] getFederatedResponses() {\n+ List<Future<FederatedResponse>> ret = new ArrayList<>();\n+ for(Pair<String, Integer> fedAddr : _fedWorkerAddresses) {\n+ InetSocketAddress isa = new InetSocketAddress(fedAddr.getLeft(), fedAddr.getRight());\n+ FederatedRequest frUDF = new FederatedRequest(RequestType.EXEC_UDF, -1,\n+ new FedStatsCollectFunction());\n+ try {\n+ ret.add(FederatedData.executeFederatedOperation(isa, frUDF));\n+ } catch(SSLException ssle) {\n+ throw new DMLRuntimeException(\"SSLException while getting the federated stats from \"\n+ + isa.toString() + \": \", ssle);\n+ } catch (Exception e) {\n+ throw new DMLRuntimeException(\"Exeption of type \" + e.getClass().getName()\n+ + \" thrown while getting stats from federated worker: \", e);\n+ }\n+ }\n+ @SuppressWarnings(\"unchecked\")\n+ Future<FederatedResponse>[] retArr = ret.toArray(new Future[0]);\n+ return retArr;\n+ }\n+\n+ private static class FedStatsCollectFunction extends FederatedUDF {\n+ private static final long serialVersionUID = 1L;\n+\n+ public FedStatsCollectFunction() {\n+ super(new long[] { });\n+ }\n+\n+ @Override\n+ public FederatedResponse execute(ExecutionContext ec, Data... data) {\n+ FedStatsCollection fedStats = new FedStatsCollection();\n+ fedStats.collectStats();\n+ return new FederatedResponse(FederatedResponse.ResponseType.SUCCESS, fedStats);\n+ }\n+\n+ @Override\n+ public Pair<String, LineageItem> getLineageItem(ExecutionContext ec) {\n+ return null;\n+ }\n+ }\n+\n+ protected static class FedStatsCollection implements Serializable {\n+ private static final long serialVersionUID = 1L;\n+\n+ private void collectStats() {\n+ cacheStats.collectStats();\n+ jitCompileTime = ((double)Statistics.getJITCompileTime()) / 1000; // in sec\n+ gcStats.collectStats();\n+ heavyHitters = Statistics.getHeavyHittersHashMap();\n+ }\n+\n+ private void aggregate(FedStatsCollection that) {\n+ cacheStats.aggregate(that.cacheStats);\n+ jitCompileTime += that.jitCompileTime;\n+ gcStats.aggregate(that.gcStats);\n+ that.heavyHitters.forEach(\n+ (key, value) -> heavyHitters.merge(key, value, (v1, v2) ->\n+ new ImmutablePair<>(v1.getLeft() + v2.getLeft(), v1.getRight() + v2.getRight()))\n+ );\n+ }\n+\n+ protected static class CacheStatsCollection implements Serializable {\n+ private static final long serialVersionUID = 1L;\n+\n+ private void collectStats() {\n+ memHits = CacheStatistics.getMemHits();\n+ linHits = CacheStatistics.getLinHits();\n+ fsBuffHits = CacheStatistics.getFSBuffHits();\n+ fsHits = CacheStatistics.getFSHits();\n+ hdfsHits = CacheStatistics.getHDFSHits();\n+ linWrites = CacheStatistics.getLinWrites();\n+ fsBuffWrites = CacheStatistics.getFSBuffWrites();\n+ fsWrites = CacheStatistics.getFSWrites();\n+ hdfsWrites = CacheStatistics.getHDFSWrites();\n+ acqRTime = ((double)CacheStatistics.getAcquireRTime()) / 1000000000; // in sec\n+ acqMTime = ((double)CacheStatistics.getAcquireMTime()) / 1000000000; // in sec\n+ rlsTime = ((double)CacheStatistics.getReleaseTime()) / 1000000000; // in sec\n+ expTime = ((double)CacheStatistics.getExportTime()) / 1000000000; // in sec\n+ }\n+\n+ private void aggregate(CacheStatsCollection that) {\n+ memHits += that.memHits;\n+ linHits += that.linHits;\n+ fsBuffHits += that.fsBuffHits;\n+ fsHits += that.fsHits;\n+ hdfsHits += that.hdfsHits;\n+ linWrites += that.linWrites;\n+ fsBuffWrites += that.fsBuffWrites;\n+ fsWrites += that.fsWrites;\n+ hdfsWrites += that.hdfsWrites;\n+ acqRTime += that.acqRTime;\n+ acqMTime += that.acqMTime;\n+ rlsTime += that.rlsTime;\n+ expTime += that.expTime;\n+ }\n+\n+ private long memHits = 0;\n+ private long linHits = 0;\n+ private long fsBuffHits = 0;\n+ private long fsHits = 0;\n+ private long hdfsHits = 0;\n+ private long linWrites = 0;\n+ private long fsBuffWrites = 0;\n+ private long fsWrites = 0;\n+ private long hdfsWrites = 0;\n+ private double acqRTime = 0;\n+ private double acqMTime = 0;\n+ private double rlsTime = 0;\n+ private double expTime = 0;\n+ }\n+\n+ protected static class GCStatsCollection implements Serializable {\n+ private static final long serialVersionUID = 1L;\n+\n+ private void collectStats() {\n+ gcCount = Statistics.getJVMgcCount();\n+ gcTime = ((double)Statistics.getJVMgcTime()) / 1000; // in sec\n+ }\n+\n+ private void aggregate(GCStatsCollection that) {\n+ gcCount += that.gcCount;\n+ gcTime += that.gcTime;\n+ }\n+\n+ private long gcCount = 0;\n+ private double gcTime = 0;\n+ }\n+\n+ private CacheStatsCollection cacheStats = new CacheStatsCollection();\n+ private double jitCompileTime = 0;\n+ private GCStatsCollection gcStats = new GCStatsCollection();\n+ private HashMap<String, Pair<Long, Double>> heavyHitters = new HashMap<>();\n+ }\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/InitFEDInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/InitFEDInstruction.java", "diff": "@@ -47,6 +47,7 @@ import org.apache.sysds.runtime.controlprogram.federated.FederatedRange;\nimport org.apache.sysds.runtime.controlprogram.federated.FederatedResponse;\nimport org.apache.sysds.runtime.controlprogram.federated.FederationMap;\nimport org.apache.sysds.runtime.controlprogram.federated.FederationMap.FType;\n+import org.apache.sysds.runtime.controlprogram.federated.FederatedStatistics;\nimport org.apache.sysds.runtime.controlprogram.federated.FederationUtils;\nimport org.apache.sysds.runtime.instructions.InstructionUtils;\nimport org.apache.sysds.runtime.instructions.cp.CPOperand;\n@@ -120,6 +121,10 @@ public class InitFEDInstruction extends FEDInstruction implements LineageTraceab\nString host = parsedValues[0];\nint port = Integer.parseInt(parsedValues[1]);\nString filePath = parsedValues[2];\n+\n+ // register the federated worker for federated statistics creation\n+ FederatedStatistics.registerFedWorker(host, port);\n+\n// get beginning and end of data ranges\nList<Data> rangesData = ranges.getData();\nData beginData = rangesData.get(i * 2);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/utils/Statistics.java", "new_path": "src/main/java/org/apache/sysds/utils/Statistics.java", "diff": "@@ -25,6 +25,7 @@ import java.lang.management.ManagementFactory;\nimport java.text.DecimalFormat;\nimport java.util.Arrays;\nimport java.util.Comparator;\n+import java.util.HashMap;\nimport java.util.List;\nimport java.util.Map.Entry;\nimport java.util.Set;\n@@ -32,12 +33,15 @@ import java.util.concurrent.ConcurrentHashMap;\nimport java.util.concurrent.atomic.DoubleAdder;\nimport java.util.concurrent.atomic.LongAdder;\n+import org.apache.commons.lang3.tuple.ImmutablePair;\n+import org.apache.commons.lang3.tuple.Pair;\nimport org.apache.sysds.api.DMLScript;\nimport org.apache.sysds.conf.ConfigurationManager;\nimport org.apache.sysds.hops.OptimizerUtils;\nimport org.apache.sysds.runtime.controlprogram.caching.CacheStatistics;\nimport org.apache.sysds.runtime.controlprogram.context.SparkExecutionContext;\nimport org.apache.sysds.runtime.controlprogram.federated.FederatedRequest.RequestType;\n+import org.apache.sysds.runtime.controlprogram.federated.FederatedStatistics;\nimport org.apache.sysds.runtime.controlprogram.parfor.stat.Timing;\nimport org.apache.sysds.runtime.instructions.Instruction;\nimport org.apache.sysds.runtime.instructions.InstructionUtils;\n@@ -732,6 +736,17 @@ public class Statistics\nreturn (tmp != null) ? tmp.count.longValue() : 0;\n}\n+ public static HashMap<String, Pair<Long, Double>> getHeavyHittersHashMap() {\n+ HashMap<String, Pair<Long, Double>> heavyHitters = new HashMap<>();\n+ for(String opcode : _instStats.keySet()) {\n+ InstStats val = _instStats.get(opcode);\n+ long count = val.count.longValue();\n+ double time = val.time.longValue() / 1000000000d; // in sec\n+ heavyHitters.put(opcode, new ImmutablePair<Long, Double>(new Long(count), new Double(time)));\n+ }\n+ return heavyHitters;\n+ }\n+\n/**\n* Obtain a string tabular representation of the heavy hitter instructions\n* that displays the time, instruction count, and optionally GPU stats about\n@@ -956,7 +971,7 @@ public class Statistics\n}\n- private static String [] wrap(String str, int wrapLength) {\n+ public static String [] wrap(String str, int wrapLength) {\nint numLines = (int) Math.ceil( ((double)str.length()) / wrapLength);\nint len = str.length();\nString [] ret = new String[numLines];\n@@ -1105,6 +1120,11 @@ public class Statistics\nif (DMLScript.CHECK_PRIVACY)\nsb.append(CheckedConstraintsLog.display());\n+ if(DMLScript.FED_STATISTICS) {\n+ sb.append(\"\\n\");\n+ sb.append(FederatedStatistics.displayFedStatistics(DMLScript.FED_STATISTICS_COUNT));\n+ }\n+\nreturn sb.toString();\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/federated/primitives/FederatedStatisticsTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/federated/primitives/FederatedStatisticsTest.java", "diff": "@@ -30,14 +30,12 @@ import org.apache.sysds.test.AutomatedTestBase;\nimport org.apache.sysds.test.TestConfiguration;\nimport org.apache.sysds.test.TestUtils;\nimport org.junit.Assert;\n-import org.junit.Ignore;\nimport org.junit.Test;\nimport org.junit.runner.RunWith;\nimport org.junit.runners.Parameterized;\n@RunWith(value = Parameterized.class)\[email protected]\n-@Ignore\npublic class FederatedStatisticsTest extends AutomatedTestBase {\nprivate final static String TEST_DIR = \"functions/federated/\";\n@@ -105,7 +103,6 @@ public class FederatedStatisticsTest extends AutomatedTestBase {\nTestConfiguration config = availableTestConfigurations.get(TEST_NAME);\nloadTestConfiguration(config);\n-\n// Run reference dml script with normal matrix\nfullDMLScriptName = HOME + TEST_NAME + \"Reference.dml\";\nprogramArgs = new String[] {\"-args\", input(\"X1\"), input(\"X2\"), input(\"Y\"), expected(\"Z\")};\n@@ -113,7 +110,7 @@ public class FederatedStatisticsTest extends AutomatedTestBase {\n// Run actual dml script with federated matrix\nfullDMLScriptName = HOME + TEST_NAME + \".dml\";\n- programArgs = new String[] {\"-stats\", \"30\", \"-nvargs\",\n+ programArgs = new String[] {\"-stats\", \"30\", \"-fedStats\", \"-nvargs\",\n\"in_X1=\" + TestUtils.federatedAddress(port1, input(\"X1\")),\n\"in_X2=\" + TestUtils.federatedAddress(port2, input(\"X2\")), \"rows=\" + rows, \"cols=\" + cols,\n\"in_Y=\" + input(\"Y\"), \"out=\" + output(\"Z\")};\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/privacy/algorithms/FederatedL2SVMTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/privacy/algorithms/FederatedL2SVMTest.java", "diff": "@@ -286,8 +286,7 @@ public class FederatedL2SVMTest extends AutomatedTestBase {\nrows = 1000; cols = 1;\nMap<String, PrivacyConstraint> privacyConstraints = new HashMap<>();\nprivacyConstraints.put(\"X1\", new PrivacyConstraint(PrivacyLevel.PrivateAggregation));\n- federatedL2SVM(Types.ExecMode.SINGLE_NODE, privacyConstraints, null, PrivacyLevel.PrivateAggregation,\n- false, null, true, DMLRuntimeException.class);\n+ federatedL2SVMNoException(Types.ExecMode.SINGLE_NODE, privacyConstraints, null, PrivacyLevel.PrivateAggregation);\n}\n@Test\n@@ -295,8 +294,7 @@ public class FederatedL2SVMTest extends AutomatedTestBase {\nrows = 1000; cols = 1;\nMap<String, PrivacyConstraint> privacyConstraints = new HashMap<>();\nprivacyConstraints.put(\"X2\", new PrivacyConstraint(PrivacyLevel.PrivateAggregation));\n- federatedL2SVM(Types.ExecMode.SINGLE_NODE, privacyConstraints, null, PrivacyLevel.PrivateAggregation,\n- false, null, true, DMLRuntimeException.class);\n+ federatedL2SVMNoException(Types.ExecMode.SINGLE_NODE, privacyConstraints, null, PrivacyLevel.PrivateAggregation);\n}\n@Test\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/builtin/cvlm.dml", "new_path": "scripts/builtin/cvlm.dml", "diff": "m_cvlm = function(Matrix[Double] X, Matrix[Double] y, Integer k, Integer icpt = 0, Double reg = 1e-7) return (Matrix[Double] y_predict, Matrix[Double] allbeta)\n{\nM = nrow(X);\n- lim = as.integer(M/k);\n+ lim = floor(as.integer(M/k));\ny_predict = y;\nallbeta = matrix(0, rows=k, cols=ncol(X));\n@@ -36,6 +36,11 @@ m_cvlm = function(Matrix[Double] X, Matrix[Double] y, Integer k, Integer icpt =\ntrainSet = X[testE+1:M,];\ntrainRes = y[testE+1:M,];\n}\n+ else if(i == k)\n+ {\n+ trainSet = X[1:testS-1,];\n+ trainRes = y[1:testS-1,];\n+ }\nelse {\ntrainSet = rbind(X[1:testS-1,], X[testE+1:M,]);\ntrainRes = rbind(y[1:testS-1,], y[testE+1:M,]);\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/builtin/correctTypos.dml", "new_path": "scripts/builtin/correctTypos.dml", "diff": "# TODO: future: add parameter for list of words that are sure to be correct\n-s_correctTypos = function(Frame[String] strings, Double frequency_threshold=0.05, Integer distance_threshold=2,\n+s_correctTypos = function(Frame[String] strings, Matrix[Double] nullMask, Double frequency_threshold=0.05, Integer distance_threshold=2,\nBoolean decapitalize=TRUE, Boolean correct=TRUE, Boolean is_verbose=FALSE)\nreturn (Frame[String] Y)\n{\nif(is_verbose)\nprint (\"BEGIN CORRECT-TYPOS SCRIPT\");\n-\nnum_strings = length(strings);\nif(is_verbose)\n@@ -67,16 +66,21 @@ s_correctTypos = function(Frame[String] strings, Double frequency_threshold=0.05\nif (decapitalize)\nstrings = map(strings, \"s -> s.toLowerCase()\");\n+\n+ if(nrow(strings) != nrow(nullMask) | ncol(strings) != ncol(nullMask))\n+ stop(\"Dimension mismatch: data dimensions do not match with mask dimensions\")\nY = strings\n# build dictionary\ncurrent_string = as.scalar(strings[1]);\n- dict = cbind(as.frame(current_string), as.frame(1));\n+ dict = cbind(as.frame(\"\"), as.frame(1));\n- for (i in 2:num_strings) {\n+ for (i in 1:num_strings) {\ncurrent_string = as.scalar(strings[i]);\n+ if(as.scalar(nullMask[i]) == 0)\ndict = insertOrIncrement(current_string, dict);\n}\n+ dict = dict[2:nrow(dict),]\nstrings = dict[,1];\nfrequencies = as.matrix(dict[,2]) / num_strings;\nlengths = as.matrix(map(strings, \"s -> s.length()\"));\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/builtin/correct_typos.dml", "new_path": "src/test/scripts/functions/builtin/correct_typos.dml", "diff": "#-------------------------------------------------------------\nX = read($X, data_type=\"frame\", format=\"csv\", header=FALSE);\n-Y = correctTypos(X, $frequency_threshold, $distance_threshold, $decapitalize, $correct, $is_verbose);\n+nullMask = matrix(0, rows=nrow(X), cols=ncol(X))\n+Y = correctTypos(X, nullMask, $frequency_threshold, $distance_threshold, $decapitalize, $correct, $is_verbose);\nwrite(Y, $Y, format=\"csv\")\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/builtin/frameSort.dml", "new_path": "scripts/builtin/frameSort.dml", "diff": "-s_frameSort = function(Frame[String] F)\n+s_frameSort = function(Frame[String] F, Boolean orderDesc = TRUE )\nreturn (Frame[String] f_odered)\n{\nidx = matrix(1, 1, ncol(F))\n@@ -47,6 +47,6 @@ return (Frame[String] f_odered)\n# recode logical pipelines for easy handling\njspecR = \"{ids:true, recode:[\"+index+\"]}\";\n[X, M] = transformencode(target=F, spec=jspecR);\n- ordered = order(target = X, by = 1, decreasing=TRUE, index.return=FALSE)\n+ ordered = order(target = X, by = 1, decreasing=orderDesc, index.return=FALSE)\nf_odered = transformdecode(target=ordered, spec=jspecR, meta=M);\n}\n\\ No newline at end of file\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/builtin/bandit.dml", "new_path": "scripts/builtin/bandit.dml", "diff": "@@ -95,7 +95,7 @@ m_bandit = function(Matrix[Double] X_train, Matrix[Double] Y_train, List[Unknown\n# sort the configurations for successive halving\navergae_perf = getMaxPerConf(outPip, nrow(configurations))\n- configurations = frameSort(cbind(avergae_perf, configurations))\n+ configurations = frameSort(cbind(avergae_perf, configurations), TRUE)\nconfigurations = configurations[, 2:ncol(configurations)]\n}\nbracket_pipel = removeEmpty(target=bracket_pipel, margin=\"rows\")\n@@ -405,7 +405,7 @@ extractTopK = function(Frame[Unknown] pipeline, Matrix[Double] hyperparam,\n# sort results\nhyperparam = order(target = hyperparam, by = 1, decreasing=TRUE, index.return=FALSE)\n- pipeline = frameSort(pipeline)\n+ pipeline = frameSort(pipeline, TRUE)\n# remove the row with accuracy less than test accuracy\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/pipelines/CleaningTestClassification.java", "new_path": "src/test/java/org/apache/sysds/test/functions/pipelines/CleaningTestClassification.java", "diff": "@@ -40,7 +40,6 @@ public class CleaningTestClassification extends AutomatedTestBase {\nprivate final static String CLEAN = DATA_DIR+ \"clean.csv\";\nprivate final static String META = RESOURCE+ \"meta/meta_census.csv\";\nprivate final static String OUTPUT = RESOURCE+\"intermediates/\";\n- private final static String LOGICAL = RESOURCE+\"intermediates/logical.csv\";\nprivate static final String PARAM_DIR = \"./scripts/pipelines/properties/\";\nprivate final static String PARAM = PARAM_DIR + \"param.csv\";\n@@ -79,7 +78,7 @@ public class CleaningTestClassification extends AutomatedTestBase {\nloadTestConfiguration(getTestConfiguration(TEST_NAME1));\nfullDMLScriptName = HOME + TEST_NAME1 + \".dml\";\nprogramArgs = new String[] {\"-stats\", \"-exec\", \"singlenode\", \"-nvargs\", \"dirtyData=\"+DIRTY,\n- \"metaData=\"+META, \"primitives=\"+PRIMITIVES, \"parameters=\"+PARAM, \"logical=\"+LOGICAL,\n+ \"metaData=\"+META, \"primitives=\"+PRIMITIVES, \"parameters=\"+PARAM,\n\"sampleSize=\"+ sample, \"topk=\"+ topk, \"rv=\"+ resources, \"cv=\"+ crossfold,\n\"weighted=\"+ weightedAccuracy, \"output=\"+OUTPUT, \"target=\"+target, \"cleanData=\"+CLEAN,\n\"O=\"+output(\"O\")};\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/intermediates/hyperparams.csv", "new_path": "src/test/scripts/functions/pipelines/intermediates/hyperparams.csv", "diff": "-36.0,0,0,0,0,1.0,0,0,0,2.0,2.0,0,1.0,0,0,0,0,0,0,0,0,0,0,1.0,0,0,0,2.0,3.0,86.0,1.0,0,0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n-36.0,0,0,0,0,1.0,0,0,0,2.0,2.0,1.0,1.0,0,0,0,0,0,0,0,0,0,0,1.0,0,0,0,2.0,3.0,67.0,1.0,0,0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n-36.0,0,0,0,0,1.0,0,0,0,2.0,2.0,1.0,0,0,0,0,0,0,0,0,0,0,0,1.0,0,0,0,2.0,3.0,102.0,1.0,0,0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n-36.0,0,0,0,0,1.0,0,0,0,2.0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,1.0,0,0,0,2.0,3.0,104.0,1.0,0,0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n-36.0,0,0,0,0,1.0,0,0,0,2.0,2.0,1.0,1.0,0,0,0,0,0,0,0,0,0,0,1.0,0,0,0,2.0,3.0,51.0,1.0,0,0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n+12.0,0,1.0,0,0,0,2.0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n+12.0,0,1.0,0,0,0,2.0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/intermediates/logical.csv", "new_path": "src/test/scripts/functions/pipelines/intermediates/logical.csv", "diff": "-MVI,OTLR,DUMMY,DIM\n+MVI,DUMMY,DIM\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/intermediates/pipelines.csv", "new_path": "src/test/scripts/functions/pipelines/intermediates/pipelines.csv", "diff": "-imputeByMean,scale,dummycoding,m_pca\n-imputeByMean,scale,dummycoding,m_pca\n-imputeByMedian,scale,dummycoding,m_pca\n-imputeByMedian,scale,dummycoding,m_pca\n-imputeByMean,scale,dummycoding,m_pca\n+imputeByMean,dummycoding\n+imputeByMedian,dummycoding\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/testClassification.dml", "new_path": "src/test/scripts/functions/pipelines/testClassification.dml", "diff": "@@ -29,7 +29,7 @@ F = read($dirtyData, data_type=\"frame\", format=\"csv\", header=TRUE,\nmetaInfo = read($metaData, data_type=\"frame\", format=\"csv\", header=FALSE);\nprimitives = read($primitives, data_type = \"frame\", format=\"csv\", header= TRUE)\nparam = read($parameters, data_type = \"frame\", format=\"csv\", header= TRUE)\n-logical = read($logical, data_type = \"frame\", format=\"csv\", header= FALSE)\n+logical = frame([\"MVI\", \"DUMMY\"], rows=1, cols=2)\nsample = $sampleSize\ntopK = $topk\nresources = $rv\n@@ -80,7 +80,8 @@ getSchema = getSchema[, 1:ncol(getSchema) - 1] # strip the mask of class label\n# 5. find the best hyper parameters for classification algorithm\n# for now only find the best values for intercept and maximum outer iteration\n-opt = utils::getOpByTarget(eX, eY, getMask, targetApplicaton)\n+opt = matrix(\"0 0 100\", rows=1, cols=3)\n+\n# 6. get the cross validated accuracy on dirty dataset (only on training set)\nd_accuracy = 0\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/testLogical.dml", "new_path": "src/test/scripts/functions/pipelines/testLogical.dml", "diff": "@@ -68,7 +68,7 @@ getMask = getMask[, 1:ncol(getMask) - 1] # strip the mask of class label\ngetFdMask = getFdMask[, 1:ncol(getFdMask) - 1] # strip the mask of class label\ngetSchema = getSchema[, 1:ncol(getSchema) - 1] # strip the mask of class label\n# hyperparam for classifier\n-opt = utils::getOpByTarget(eX, eY, getMask, targetApplicaton)\n+opt = matrix(\"0 0 100\", rows=1, cols=3)\nprint(\"opt \"+toString(opt))\n# get the cross validated accuracy on dirty dataset (only on training set)\nd_accuracy = 0\n" } ]

[ { "change_type": "MODIFY", "old_path": ".github/workflows/applicationTests.yml", "new_path": ".github/workflows/applicationTests.yml", "diff": "@@ -26,6 +26,9 @@ on:\nbranches:\n- master\npull_request:\n+ paths-ignore:\n+ - 'docs/**'\n+ - '*.md'\nbranches:\n- master\n" }, { "change_type": "MODIFY", "old_path": ".github/workflows/build.yml", "new_path": ".github/workflows/build.yml", "diff": "name: Build\n-on: [push, pull_request]\n-\n+on:\n+ push:\n+ branches:\n+ - master\n+ pull_request:\n+ paths-ignore:\n+ - 'docs/**'\n+ - '*.md'\n+ branches:\n+ - master\njobs:\nbuild:\nruns-on: ${{ matrix.os }}\n" }, { "change_type": "MODIFY", "old_path": ".github/workflows/componentTests.yml", "new_path": ".github/workflows/componentTests.yml", "diff": "name: Component Test\n-on: [push, pull_request]\n+on:\n+ push:\n+ branches:\n+ - master\n+ pull_request:\n+ paths-ignore:\n+ - 'docs/**'\n+ - '*.md'\n+ branches:\n+ - master\njobs:\ncomponentTests:\n" }, { "change_type": "MODIFY", "old_path": ".github/workflows/documentation.yml", "new_path": ".github/workflows/documentation.yml", "diff": "name: Documentation\n-on: [push, pull_request]\n+on:\n+ push:\n+ branches:\n+ - master\n+ pull_request:\n+ paths-ignore:\n+ - 'docs/**'\n+ - '*.md'\n+ branches:\n+ - master\njobs:\ndocumentation1:\n" }, { "change_type": "MODIFY", "old_path": ".github/workflows/federatedPython.yml", "new_path": ".github/workflows/federatedPython.yml", "diff": "@@ -26,6 +26,9 @@ on:\nbranches:\n- master\npull_request:\n+ paths-ignore:\n+ - 'docs/**'\n+ - '*.md'\nbranches:\n- master\n" }, { "change_type": "MODIFY", "old_path": ".github/workflows/functionsTests.yml", "new_path": ".github/workflows/functionsTests.yml", "diff": "@@ -26,6 +26,9 @@ on:\nbranches:\n- master\npull_request:\n+ paths-ignore:\n+ - 'docs/**'\n+ - '*.md'\nbranches:\n- master\n" }, { "change_type": "MODIFY", "old_path": ".github/workflows/python.yml", "new_path": ".github/workflows/python.yml", "diff": "@@ -26,6 +26,9 @@ on:\nbranches:\n- master\npull_request:\n+ paths-ignore:\n+ - 'docs/**'\n+ - '*.md'\nbranches:\n- master\n" } ]

[ { "change_type": "MODIFY", "old_path": "docs/_includes/header.html", "new_path": "docs/_includes/header.html", "diff": "@@ -44,6 +44,7 @@ limitations under the License.\n<li><b>Running SystemDS:</b></li>\n<li><a href=\".{% if page.path contains 'site' %}/..{% endif %}/site/run\">Standalone Guide</a></li>\n<li><a href=\".{% if page.path contains 'site' %}/..{% endif %}/site/gpu\">GPU Guide</a></li>\n+ <li><a href=\".{% if page.path contains 'site' %}/..{% endif %}/site/native-backend\">Native Backend (BLAS)</a></li>\n<li class=\"divider\"></li>\n<li><b>Language Guides:</b></li>\n<li><a href=\".{% if page.path contains 'site' %}/..{% endif %}/site/dml-language-reference.html\">DML Language Reference</a></li>\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/docs/source/conf.py", "new_path": "src/main/python/docs/source/conf.py", "diff": "@@ -38,7 +38,7 @@ copyright = '2021, Apache SystemDS'\nauthor = 'Apache SystemDS'\n# The full version, including alpha/beta/rc tags\n-release = '2.0.0'\n+release = '2.2.0-SNAPSHOT'\n# -- General configuration ---------------------------------------------------\n# Add any Sphinx extension module names here, as strings.\n" } ]

[ { "change_type": "MODIFY", "old_path": ".github/workflows/python.yml", "new_path": ".github/workflows/python.yml", "diff": "@@ -84,9 +84,6 @@ jobs:\npython-version: ${{ matrix.python-version }}\narchitecture: 'x64'\n- - name: Install Protobuf\n- run: sudo apt-get install protobuf-compiler libprotoc-dev\n-\n- name: Install pip Dependencies\nrun: pip install numpy py4j wheel scipy sklearn requests pandas\n" } ]

[ { "change_type": "MODIFY", "old_path": "docker/README.md", "new_path": "docker/README.md", "diff": "@@ -36,22 +36,31 @@ To execute any given DML script follow the step Run.\n## Run\n-Running SystemDS in a docker container is as simple as constructing any DML script\n-Then Download the docker image `systemds/sysds:latest` or build your own.\n+Running SystemDS in a docker container is as simple as constructing any DML script.\n+Then download the docker image `systemds/sysds:latest` or build your own.\n```bash\ndocker pull systemds/sysds:latest\n```\n-Verify that the docker image correctly works simply by running it, make sure that your terminal is pointing at the root of you systemds git clone.\n+Verify that the docker image correctly works simply by running it.\n```bash\n-./docker/runDocker.sh\n+docker run --rm systemds/sysds:latest\n```\n-This above command will mount the folder `docker/mountFolder`, and execute the script named main.dml inside the folder using the docker container.\n+It should respond with something like:\n-You can mount any such folder and execute systemds on by changing the first part of the -v argument of the following command:\n+```txt\n+Hello, World!\n+SystemDS Statistics:\n+Total execution time: 0.010 sec.\n+```\n+\n+To run specific scripts mount a folder(s) containing the scripts and data,\n+and execute the script inside the folder using the docker container.\n+\n+You can mount any such folder and execute systemds by changing the first part of the -v argument of the following command:\n```bash\ndocker run \\\n@@ -59,7 +68,54 @@ docker run \\\n--rm systemds/sysds:latest\n```\n-## Testing\n+Default behavior is to run the script located at /input/main.dml.\n+To run any other script use:\n+\n+```bash\n+docker run \\\n+ -v $(pwd)/folder/to/share:/any/path/in/docker/instance \\\n+ --rm systemds/sysds:latest \\\n+ systemds /any/path/to/a/script.dml\n+```\n+\n+### Docker run worker node\n+\n+To run a federated worker in a docker container simply use:\n+\n+```bash\n+docker run -p 8000:8000 --rm systemds/sysds:latest systemds WORKER 8000\n+```\n+\n+This port forwards the worker to port 8000 on the host and starts a worker in the instance on port 8000.\n+\n+Note that the worker does not have any data, since no data is mounted in the worker image.\n+To add a folder containing the data needed in the worker do the following:\n+\n+```bash\n+docker run \\\n+ -p 8000:8000 \\\n+ -v $(pwd)/data/folder/path/locally:/data/folder/path/in/container \\\n+ --rm systemds/sysds:latest systemds WORKER 8000\n+```\n+\n+### Docker run python script\n+\n+To run a python script the `pythonsysds` image is used.\n+\n+```bash\n+docker run --rm systemds/pythonsysds:latest\n+```\n+\n+User provided scripts have to be mounted into the image.\n+\n+```bash\n+docker run \\\n+ -v $(pwd)/data/folder/path/locally:/data/folder/path/in/container \\\n+ --rm systemds/pythonsystds:latest \\\n+ python3 path/to/script/to/execute.py\n+```\n+\n+## Testing image\nWe also have a docker image for execution of tests.\nThis enables faster test execution on the github actions.\n@@ -83,5 +139,5 @@ Test your testing image locally by running the following command:\ndocker run \\\n-v $(pwd):/github/workspace \\\nsystemds/testingsysds:latest \\\n- org.apache.sysds.test.component.*.**\n+ org.apache.sysds.test.component.**\n```\n" }, { "change_type": "RENAME", "old_path": "docker/runDocker.sh", "new_path": "docker/mountFolder/main.py", "diff": "-#/bin/bash\n#-------------------------------------------------------------\n#\n# Licensed to the Apache Software Foundation (ASF) under one\n#\n#-------------------------------------------------------------\n-# The image is on docker so it is posible to run from here.\n-# Execute the docker container\n-docker run \\\n- -v $(pwd)/docker/mountFolder:/input \\\n- --rm systemds/sysds:latest\n+from systemds.context import SystemDSContext\n+\n+with SystemDSContext() as sds:\n+ # Make a script (lazy evaluation nothing happens):\n+ script = sds.scalar(\"Hello World\").print()\n+ # Compute script to execute:\n+ script.compute()\n+ # Print the stdout from the context:\n+ print(sds.get_stdout()[0])\n+\n\\ No newline at end of file\n" }, { "change_type": "MODIFY", "old_path": "docker/pythonsysds.Dockerfile", "new_path": "docker/pythonsysds.Dockerfile", "diff": "#\n#-------------------------------------------------------------\n-FROM openjdk:8-alpine\n+FROM ubuntu:20.04\n-RUN apk add --no-cache --virtual .build-deps g++ python3-dev libffi-dev openssl-dev && \\\n- apk add --no-cache python3 && \\\n- pip3 install --upgrade pip setuptools\n+WORKDIR /usr/src/\n-RUN pip3 install systemds\n+# Maven\n+ENV MAVEN_VERSION 3.6.3\n+ENV MAVEN_HOME /usr/lib/mvn\n+ENV PATH $MAVEN_HOME/bin:$PATH\n+# Java\n+ENV JAVA_HOME /usr/lib/jvm/java-8-openjdk-amd64\n+ENV PATH $JAVA_HOME/bin:$MAVEN_HOME/bin:$PATH\n+ENV SYSTEMDS_ROOT=/usr/src/systemds\n+ENV PATH $SYSTEMDS_ROOT/bin:$PATH\n+ENV SYSDS_QUIET=1\n+\n+RUN apt-get update -qq \\\n+ && apt-get upgrade -y \\\n+ && apt-get install -y --no-install-recommends \\\n+ wget \\\n+ git \\\n+ ca-certificates \\\n+ && apt-get clean \\\n+ && mkdir -p /usr/lib/jvm \\\n+ && wget -qO- \\\n+ https://github.com/AdoptOpenJDK/openjdk8-binaries/releases/download/jdk8u282-b08/OpenJDK8U-jdk_x64_linux_hotspot_8u282b08.tar.gz | tar xzf - \\\n+ && mv jdk8u282-b08 /usr/lib/jvm/java-8-openjdk-amd64 \\\n+ && wget -qO- \\\n+ http://archive.apache.org/dist/maven/maven-3/$MAVEN_VERSION/binaries/apache-maven-$MAVEN_VERSION-bin.tar.gz | tar xzf - \\\n+ && mv apache-maven-$MAVEN_VERSION /usr/lib/mvn \\\n+ && git clone --depth 1 https://github.com/apache/systemds.git systemds && \\\n+ cd /usr/src/systemds/ && \\\n+ mvn clean package -P distribution && \\\n+ cd /usr/src/systemds/src/main/python && \\\n+ apt-get install -y --no-install-recommends \\\n+ python3 python3-pip && \\\n+ apt-get clean && \\\n+ python3 -m pip install --upgrade pip && \\\n+ pip3 install numpy py4j wheel requests pandas && \\\n+ python3 create_python_dist.py && \\\n+ pip3 install . && \\\n+ cd /usr/src/systemds/ && \\\n+ rm -r docker && \\\n+ rm -r docs && \\\n+ rm -r src && \\\n+ rm -r /usr/lib/mvn && \\\n+ rm -r CONTRIBUTING.md && \\\n+ rm -r pom.xml && \\\n+ rm -r ~/.m2 && \\\n+ rm -r .github && \\\n+ rm -r dev && \\\n+ rm -r .git\n+\n+COPY docker/mountFolder/main.py /input/main.py\n+\n+CMD [\"python3\", \"/input/main.py\"]\n" }, { "change_type": "MODIFY", "old_path": "docker/sysds.Dockerfile", "new_path": "docker/sysds.Dockerfile", "diff": "#\n#-------------------------------------------------------------\n-# Use Alpine OpenJDK 8 base\n-FROM openjdk:8-alpine\n+FROM ubuntu:20.04\nWORKDIR /usr/src/\n-# Install Maven\n-# Credit https://github.com/Zenika/alpine-maven/blob/master/jdk8/Dockerfile\n-\n+# Maven\nENV MAVEN_VERSION 3.6.3\nENV MAVEN_HOME /usr/lib/mvn\nENV PATH $MAVEN_HOME/bin:$PATH\n-\n-RUN wget http://archive.apache.org/dist/maven/maven-3/$MAVEN_VERSION/binaries/apache-maven-$MAVEN_VERSION-bin.tar.gz && \\\n- tar -zxvf apache-maven-$MAVEN_VERSION-bin.tar.gz && \\\n- rm apache-maven-$MAVEN_VERSION-bin.tar.gz && \\\n- mv apache-maven-$MAVEN_VERSION /usr/lib/mvn\n-\n-# Install Extras\n-RUN apk add --no-cache git bash\n-\n-RUN git clone https://github.com/apache/systemds.git systemds\n-\n-WORKDIR /usr/src/systemds/\n-\n-RUN mvn clean package -P distribution\n-\n-# Remove Maven since it is not needed for running the system\n-RUN rm -r /usr/lib/mvn\n-\n+# Java\n+ENV JAVA_HOME /usr/lib/jvm/java-8-openjdk-amd64\n+ENV PATH $JAVA_HOME/bin:$MAVEN_HOME/bin:$PATH\nENV SYSTEMDS_ROOT=/usr/src/systemds\nENV PATH $SYSTEMDS_ROOT/bin:$PATH\n+ENV SYSDS_QUIET=1\n-# Remove extra files.\n-RUN rm -r src/ && \\\n- rm -r .git\n+RUN apt-get update -qq \\\n+ && apt-get upgrade -y \\\n+ && apt-get install -y --no-install-recommends \\\n+ wget \\\n+ git \\\n+ ca-certificates \\\n+ && apt-get clean \\\n+ && mkdir -p /usr/lib/jvm \\\n+ && wget -qO- \\\n+ https://github.com/AdoptOpenJDK/openjdk8-binaries/releases/download/jdk8u282-b08/OpenJDK8U-jdk_x64_linux_hotspot_8u282b08.tar.gz | tar xzf - \\\n+ && mv jdk8u282-b08 /usr/lib/jvm/java-8-openjdk-amd64 \\\n+ && wget -qO- \\\n+ http://archive.apache.org/dist/maven/maven-3/$MAVEN_VERSION/binaries/apache-maven-$MAVEN_VERSION-bin.tar.gz | tar xzf - \\\n+ && mv apache-maven-$MAVEN_VERSION /usr/lib/mvn \\\n+ && git clone --depth 1 https://github.com/apache/systemds.git systemds && \\\n+ cd /usr/src/systemds/ && \\\n+ mvn clean package -P distribution && \\\n+ rm -r docker && \\\n+ rm -r docs && \\\n+ rm -r src && \\\n+ rm -r /usr/lib/mvn && \\\n+ rm -r CONTRIBUTING.md && \\\n+ rm -r pom.xml && \\\n+ rm -r ~/.m2\nCOPY docker/mountFolder/main.dml /input/main.dml\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/parser/ParameterizedBuiltinFunctionExpression.java", "new_path": "src/main/java/org/apache/sysds/parser/ParameterizedBuiltinFunctionExpression.java", "diff": "@@ -478,7 +478,9 @@ public class ParameterizedBuiltinFunctionExpression extends DataIdentifier\nprivate void validateReplace(DataIdentifier output, boolean conditional) {\n//check existence and correctness of arguments\nExpression target = getVarParam(\"target\");\n+ if( target.getOutput().getDataType() != DataType.FRAME ){\ncheckTargetParam(target, conditional);\n+ }\nExpression pattern = getVarParam(\"pattern\");\nif( pattern==null ) {\n@@ -497,7 +499,10 @@ public class ParameterizedBuiltinFunctionExpression extends DataIdentifier\n}\n// Output is a matrix with same dims as input\n- output.setDataType(DataType.MATRIX);\n+ output.setDataType(target.getOutput().getDataType());\n+ if(target.getOutput().getDataType() == DataType.FRAME)\n+ output.setValueType(ValueType.STRING);\n+ else\noutput.setValueType(ValueType.FP64);\noutput.setDimensions(target.getOutput().getDim1(), target.getOutput().getDim2());\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/ParameterizedBuiltinCPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/ParameterizedBuiltinCPInstruction.java", "diff": "@@ -225,6 +225,14 @@ public class ParameterizedBuiltinCPInstruction extends ComputationCPInstruction\nec.releaseMatrixInput(params.get(\"select\"));\n}\nelse if(opcode.equalsIgnoreCase(\"replace\")) {\n+ if(ec.isFrameObject(params.get(\"target\"))){\n+ FrameBlock target = ec.getFrameInput(params.get(\"target\"));\n+ String pattern = params.get(\"pattern\");\n+ String replacement = params.get(\"replacement\");\n+ FrameBlock ret = target.replaceOperations(pattern, replacement);\n+ ec.setFrameOutput(output.getName(), ret);\n+ ec.releaseFrameInput(params.get(\"target\"));\n+ }else{\nMatrixBlock target = ec.getMatrixInput(params.get(\"target\"));\ndouble pattern = Double.parseDouble(params.get(\"pattern\"));\ndouble replacement = Double.parseDouble(params.get(\"replacement\"));\n@@ -232,6 +240,8 @@ public class ParameterizedBuiltinCPInstruction extends ComputationCPInstruction\nec.setMatrixOutput(output.getName(), ret);\nec.releaseMatrixInput(params.get(\"target\"));\n}\n+\n+ }\nelse if(opcode.equals(\"lowertri\") || opcode.equals(\"uppertri\")) {\nMatrixBlock target = ec.getMatrixInput(params.get(\"target\"));\nboolean lower = opcode.equals(\"lowertri\");\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/ParameterizedBuiltinSPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/ParameterizedBuiltinSPInstruction.java", "diff": "@@ -358,6 +358,18 @@ public class ParameterizedBuiltinSPInstruction extends ComputationSPInstruction\n}\n}\nelse if(opcode.equalsIgnoreCase(\"replace\")) {\n+ if(sec.isFrameObject(params.get(\"target\"))){\n+ params.get(\"target\");\n+ JavaPairRDD<Long, FrameBlock> in1 = sec.getFrameBinaryBlockRDDHandleForVariable(params.get(\"target\"));\n+ DataCharacteristics mcIn = sec.getDataCharacteristics(params.get(\"target\"));\n+ String pattern = params.get(\"pattern\");\n+ String replacement = params.get(\"replacement\");\n+ JavaPairRDD<Long, FrameBlock> out = in1.mapValues(new RDDFrameReplaceFunction(pattern, replacement));\n+ sec.setRDDHandleForVariable(output.getName(), out);\n+ sec.addLineageRDD(output.getName(), params.get(\"target\"));\n+ sec.getDataCharacteristics(output.getName()).set(mcIn.getRows(), mcIn.getCols(), mcIn.getBlocksize(), mcIn.getNonZeros());\n+ }\n+ else {\nJavaPairRDD<MatrixIndexes, MatrixBlock> in1 = sec\n.getBinaryMatrixBlockRDDHandleForVariable(params.get(\"target\"));\nDataCharacteristics mcIn = sec.getDataCharacteristics(params.get(\"target\"));\n@@ -372,12 +384,13 @@ public class ParameterizedBuiltinSPInstruction extends ComputationSPInstruction\nsec.addLineageRDD(output.getName(), params.get(\"target\"));\n// update output statistics (required for correctness)\n- DataCharacteristics mcOut = sec.getDataCharacteristics(output.getName());\n- mcOut.set(mcIn.getRows(),\n+ sec.getDataCharacteristics(output.getName()).set(mcIn.getRows(),\nmcIn.getCols(),\nmcIn.getBlocksize(),\n(pattern != 0 && replacement != 0) ? mcIn.getNonZeros() : -1);\n}\n+\n+ }\nelse if(opcode.equalsIgnoreCase(\"lowertri\") || opcode.equalsIgnoreCase(\"uppertri\")) {\nJavaPairRDD<MatrixIndexes, MatrixBlock> in1 = sec\n.getBinaryMatrixBlockRDDHandleForVariable(params.get(\"target\"));\n@@ -544,6 +557,22 @@ public class ParameterizedBuiltinSPInstruction extends ComputationSPInstruction\n}\n}\n+ public static class RDDFrameReplaceFunction implements Function<FrameBlock, FrameBlock>{\n+ private static final long serialVersionUID = 6576713401901671660L;\n+ private final String _pattern;\n+ private final String _replacement;\n+\n+ public RDDFrameReplaceFunction(String pattern, String replacement){\n+ _pattern = pattern;\n+ _replacement = replacement;\n+ }\n+\n+ @Override\n+ public FrameBlock call(FrameBlock arg0){\n+ return arg0.replaceOperations(_pattern, _replacement);\n+ }\n+ }\n+\nprivate static class RDDExtractTriangularFunction\nimplements PairFlatMapFunction<Iterator<Tuple2<MatrixIndexes, MatrixBlock>>, MatrixIndexes, MatrixBlock> {\nprivate static final long serialVersionUID = 2754868819184155702L;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/matrix/data/FrameBlock.java", "new_path": "src/main/java/org/apache/sysds/runtime/matrix/data/FrameBlock.java", "diff": "@@ -2237,4 +2237,17 @@ public class FrameBlock implements CacheBlock, Externalizable {\npublic String apply(String input) {return null;}\npublic String apply(String input1, String input2) { return null;}\n}\n+\n+ public FrameBlock replaceOperations(String pattern, String replacement){\n+ FrameBlock ret = new FrameBlock(this);\n+ for(int i = 0; i < ret.getNumColumns(); i++){\n+ Array colData = ret._coldata[i];\n+ for(int j = 0; j < colData._size; j++){\n+ Object ent = colData.get(j);\n+ if(ent != null && ent.equals(pattern))\n+ colData.set(j,replacement);\n+ }\n+ }\n+ return ret;\n+ }\n}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysds/test/functions/frame/FrameReplaceTest.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysds.test.functions.frame;\n+\n+import static org.junit.Assert.assertTrue;\n+\n+import org.apache.sysds.api.DMLScript;\n+import org.apache.sysds.common.Types.ExecMode;\n+import org.apache.sysds.common.Types.ExecType;\n+import org.apache.sysds.test.AutomatedTestBase;\n+import org.apache.sysds.test.TestConfiguration;\n+import org.apache.sysds.test.TestUtils;\n+import org.junit.Test;\n+\n+public class FrameReplaceTest extends AutomatedTestBase {\n+ // private static final Log LOG = LogFactory.getLog(FrameReplaceTest.class.getName());\n+ private final static String TEST_DIR = \"functions/frame/\";\n+ private final static String TEST_NAME = \"ReplaceTest\";\n+ private final static String TEST_CLASS_DIR = TEST_DIR + FrameReplaceTest.class.getSimpleName() + \"/\";\n+\n+ @Override\n+ public void setUp() {\n+ TestUtils.clearAssertionInformation();\n+ addTestConfiguration(TEST_NAME, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME));\n+ }\n+\n+ @Test\n+ public void testParforFrameIntermediatesCP() {\n+ runReplaceTest(ExecType.CP);\n+ }\n+\n+ @Test\n+ public void testParforFrameIntermediatesSpark() {\n+ runReplaceTest(ExecType.SPARK);\n+ }\n+\n+ private void runReplaceTest(ExecType et) {\n+ ExecMode platformOld = rtplatform;\n+ switch(et) {\n+ case SPARK:\n+ rtplatform = ExecMode.SPARK;\n+ break;\n+ default:\n+ rtplatform = ExecMode.HYBRID;\n+ break;\n+ }\n+\n+ boolean sparkConfigOld = DMLScript.USE_LOCAL_SPARK_CONFIG;\n+ if(rtplatform == ExecMode.SPARK || rtplatform == ExecMode.HYBRID)\n+ DMLScript.USE_LOCAL_SPARK_CONFIG = true;\n+\n+ try {\n+ // setup testcase\n+ getAndLoadTestConfiguration(TEST_NAME);\n+ String HOME = SCRIPT_DIR + TEST_DIR;\n+ fullDMLScriptName = HOME + TEST_NAME + \".dml\";\n+ programArgs = new String[] {};\n+\n+ // run test\n+ String out = runTest(null).toString();\n+\n+ assertTrue(out.contains(\"south\"));\n+ assertTrue(!out.contains(\"north\"));\n+\n+ }\n+ catch(Exception ex) {\n+ throw new RuntimeException(ex);\n+ }\n+ finally {\n+ rtplatform = platformOld;\n+ DMLScript.USE_LOCAL_SPARK_CONFIG = sparkConfigOld;\n+ }\n+ }\n+\n+}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/frame/ReplaceTest.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+X = read(\"src/test/resources/datasets/homes/homes.csv\")\n+\n+X = replace(target = X, pattern=\"north\", replacement=\"south\")\n+X = replace(target = X, pattern=\"east\", replacement=\"south\")\n+X = replace(target = X, pattern=\"west\", replacement=\"south\")\n+\n+print(toString(X))\n\\ No newline at end of file\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/systemds/operator/algorithm/__init__.py", "new_path": "src/main/python/systemds/operator/algorithm/__init__.py", "diff": "@@ -98,6 +98,7 @@ from .builtin.scale import scale\nfrom .builtin.scaleApply import scaleApply\nfrom .builtin.sherlock import sherlock\nfrom .builtin.sherlockPredict import sherlockPredict\n+from .builtin.shortestPath import shortestPath\nfrom .builtin.sigmoid import sigmoid\nfrom .builtin.slicefinder import slicefinder\nfrom .builtin.smote import smote\n@@ -192,6 +193,7 @@ __all__ = ['abstain',\n'scaleApply',\n'sherlock',\n'sherlockPredict',\n+ 'shortestPath',\n'sigmoid',\n'slicefinder',\n'smote',\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/python/systemds/operator/algorithm/builtin/shortestPath.py", "diff": "+# -------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+# -------------------------------------------------------------\n+\n+# Autogenerated By : src/main/python/generator/generator.py\n+# Autogenerated From : scripts/builtin/shortestPath.dml\n+\n+from typing import Dict, Iterable\n+\n+from systemds.operator import OperationNode, Matrix, Frame, List, MultiReturn, Scalar\n+from systemds.script_building.dag import OutputType\n+from systemds.utils.consts import VALID_INPUT_TYPES\n+\n+\n+def shortestPath(G: Matrix,\n+ sourceNode: int,\n+ **kwargs: Dict[str, VALID_INPUT_TYPES]):\n+\n+ params_dict = {'G': G, 'sourceNode': sourceNode}\n+ params_dict.update(kwargs)\n+ return Matrix(G.sds_context,\n+ 'shortestPath',\n+ named_input_nodes=params_dict)\n" }, { "change_type": "MODIFY", "old_path": "src/main/python/systemds/operator/nodes/frame.py", "new_path": "src/main/python/systemds/operator/nodes/frame.py", "diff": "@@ -120,7 +120,15 @@ class Frame(OperationNode):\n\"\"\"\nColumn-wise frame concatenation, by concatenating the second frame as additional columns to the first frame.\n:param: The other frame to bind to the right hand side.\n- :return: The OperationNode containing the concatenated frames.\n+ :return: The Frame containing the concatenated frames.\n\"\"\"\nreturn Frame(self.sds_context, \"cbind\", [self, other])\n+ def replace(self, pattern:str, replacement:str) -> 'Frame':\n+ \"\"\"\n+ Replace all instances of string with replacement string\n+ :param: pattern the string to replace\n+ :param: replacement the string to replace with\n+ :return: The Frame containing the replaced values\n+ \"\"\"\n+ return Frame(self.sds_context, \"replace\", named_input_nodes={\"target\": self, \"pattern\": f\"'{pattern}'\", \"replacement\":f\"'{replacement}'\"})\n" }, { "change_type": "MODIFY", "old_path": "src/main/python/systemds/operator/nodes/matrix.py", "new_path": "src/main/python/systemds/operator/nodes/matrix.py", "diff": "@@ -348,3 +348,15 @@ class Matrix(OperationNode):\n\"\"\"\nreturn Matrix(self.sds_context, 'rev', [self])\n+ def round(self) -> 'Matrix':\n+ \"\"\" round all values to nearest natural number\n+\n+ :return: The Matrix representing the result of this operation\n+ \"\"\"\n+ return Matrix(self.sds_context, \"round\", [self])\n+\n+ def replace(self, pattern:VALID_INPUT_TYPES, replacement:VALID_INPUT_TYPES) -> 'Matrix':\n+ \"\"\"\n+ Replace all values with replacement value\n+ \"\"\"\n+ return Matrix(self.sds_context, \"replace\", named_input_nodes={\"target\": self, \"pattern\": pattern, \"replacement\":replacement})\n\\ No newline at end of file\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/python/tests/frame/test_replace.py", "diff": "+# -------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+# -------------------------------------------------------------\n+\n+import json\n+import os\n+import shutil\n+import sys\n+import unittest\n+\n+import numpy as np\n+import pandas as pd\n+from systemds.context import SystemDSContext\n+\n+\n+class TestReplaceFrame(unittest.TestCase):\n+\n+ sds: SystemDSContext = None\n+ HOMES_PATH = \"../../test/resources/datasets/homes/homes.csv\"\n+ HOMES_SCHEMA = '\"int,string,int,int,double,int,boolean,int,int\"'\n+ JSPEC_PATH = \"../../test/resources/datasets/homes/homes.tfspec_bin2.json\"\n+\n+ @classmethod\n+ def setUpClass(cls):\n+ cls.sds = SystemDSContext()\n+\n+ @classmethod\n+ def tearDownClass(cls):\n+ cls.sds.close()\n+\n+ def tearDown(self):\n+ pass\n+\n+ def test_apply_recode_bin(self):\n+\n+ F1 = self.sds.read(\n+ self.HOMES_PATH,\n+ data_type=\"frame\",\n+ schema=self.HOMES_SCHEMA,\n+ format=\"csv\",\n+ header=True,\n+ )\n+ ret = F1.replace(\"north\", \"south\").replace(\"west\", \"south\").replace(\"east\",\"south\").compute()\n+ self.assertTrue(any(ret.district == \"south\"))\n+ self.assertTrue(not( any(ret.district == \"north\")))\n+\n+\n+\n+\n+if __name__ == \"__main__\":\n+ unittest.main(exit=False)\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/python/tests/matrix/test_replace.py", "diff": "+# -------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+# -------------------------------------------------------------\n+\n+import json\n+import os\n+import random\n+import shutil\n+import sys\n+import unittest\n+\n+import numpy as np\n+import pandas as pd\n+from systemds.context import SystemDSContext\n+\n+np.random.seed(7)\n+shape = (25, 25)\n+\n+\n+class TestReplaceMatrix(unittest.TestCase):\n+\n+ sds: SystemDSContext = None\n+\n+ @classmethod\n+ def setUpClass(cls):\n+ cls.sds = SystemDSContext()\n+\n+ @classmethod\n+ def tearDownClass(cls):\n+ cls.sds.close()\n+\n+ def tearDown(self):\n+ pass\n+\n+ def test_replace_01(self):\n+ m = self.sds.rand(min=0, max=2, rows=shape[0], cols=shape[1], seed=14)\\\n+ .round().replace(1, 2).compute()\n+ self.assertTrue(1 not in m)\n+ self.assertTrue(2 in m)\n+ self.assertTrue(0 in m)\n+\n+\n+if __name__ == \"__main__\":\n+ unittest.main(exit=False)\n" } ]

[ { "change_type": "DELETE", "old_path": ".github/workflows/federatedPython.yml", "new_path": null, "diff": "-#-------------------------------------------------------------\n-#\n-# Licensed to the Apache Software Foundation (ASF) under one\n-# or more contributor license agreements. See the NOTICE file\n-# distributed with this work for additional information\n-# regarding copyright ownership. The ASF licenses this file\n-# to you under the Apache License, Version 2.0 (the\n-# \"License\"); you may not use this file except in compliance\n-# with the License. You may obtain a copy of the License at\n-#\n-# http://www.apache.org/licenses/LICENSE-2.0\n-#\n-# Unless required by applicable law or agreed to in writing,\n-# software distributed under the License is distributed on an\n-# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n-# KIND, either express or implied. See the License for the\n-# specific language governing permissions and limitations\n-# under the License.\n-#\n-#-------------------------------------------------------------\n-\n-name: Federated Python Test\n-\n-on:\n- push:\n- branches:\n- - master\n- pull_request:\n- paths-ignore:\n- - 'docs/**'\n- - '*.md'\n- branches:\n- - master\n-\n-jobs:\n- applicationsTests:\n- runs-on: ${{ matrix.os }}\n- strategy:\n- fail-fast: false\n- matrix:\n- python-version: [3.6]\n- os: [ubuntu-latest]\n- java: [ 1.8 ]\n- name: Python Test\n- steps:\n- - name: Checkout Repository\n- uses: actions/checkout@v2\n-\n- - name: Setup Java\n- uses: actions/setup-java@v1\n- with:\n- java-version: ${{ matrix.java }}\n-\n- - name: Cache Maven Dependencies\n- uses: actions/cache@v1\n- with:\n- path: ~/.m2/repository\n- key: ${{ runner.os }}-maven-${{ hashFiles('**/pom.xml') }}\n- restore-keys: |\n- ${{ runner.os }}-maven-\n-\n- - name: Maven clean & package\n- run: mvn -ntp clean package -P distribution\n-\n- - name: Setup Python\n- uses: actions/setup-python@v1\n- with:\n- python-version: ${{ matrix.python-version }}\n- architecture: 'x64'\n-\n- - name: Cache Pip Dependencies\n- uses: actions/cache@v1\n- with:\n- path: ~/.cache/pip\n- key: ${{ runner.os }}-pip-${{ matrix.python-version }}-${{ hashFiles('src/main/python/setup.py') }}\n- restore-keys: |\n- ${{ runner.os }}-pip-${{ matrix.python-version }}-\n-\n- - name: Install pip Dependencies\n- run: pip install numpy py4j wheel pandas\n-\n- - name: Build Python Package\n- run: |\n- cd src/main/python\n- python create_python_dist.py\n-\n- - name: Run Federated Python Tests\n- run: |\n- export SYSTEMDS_ROOT=$(pwd)\n- export PATH=$SYSTEMDS_ROOT/bin:$PATH\n- cd src/main/python\n- ./tests/federated/runFedTest.sh\n-\n" }, { "change_type": "MODIFY", "old_path": ".github/workflows/python.yml", "new_path": ".github/workflows/python.yml", "diff": "@@ -38,7 +38,7 @@ jobs:\nstrategy:\nfail-fast: false\nmatrix:\n- python-version: [3.6, 3.7, 3.8]\n+ python-version: [3.8]\nos: [ubuntu-latest]\njava: [ 1.8 ]\nname: Python Test\n@@ -106,3 +106,10 @@ jobs:\ncd src/main/python\npython -m unittest discover -s tests -p 'test_*.py'\necho \"Exit Status: \" $?\n+\n+ - name: Run Federated Python Tests\n+ run: |\n+ export SYSTEMDS_ROOT=$(pwd)\n+ export PATH=$SYSTEMDS_ROOT/bin:$PATH\n+ cd src/main/python\n+ ./tests/federated/runFedTest.sh\n\\ No newline at end of file\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/meta/MetaDataAll.java", "new_path": "src/main/java/org/apache/sysds/runtime/meta/MetaDataAll.java", "diff": "@@ -166,7 +166,7 @@ public class MetaDataAll extends DataIdentifier {\ncase DataExpression.READROWPARAM: _dim1 = (Integer) val; break;\ncase DataExpression.READCOLPARAM: _dim2 = (Integer) val; break;\ncase DataExpression.ROWBLOCKCOUNTPARAM: setBlocksize((Integer) val); break;\n- case DataExpression.READNNZPARAM: setNnz((Integer) val); break;\n+ case DataExpression.READNNZPARAM: setNnz(val instanceof Long ? (Long) val : (Integer) val); break;\ncase DataExpression.FORMAT_TYPE: setFormatTypeString((String) val); break;\ncase DataExpression.DATATYPEPARAM: setDataType(Types.DataType.valueOf(((String) val).toUpperCase())); break;\ncase DataExpression.VALUETYPEPARAM: setValueType(Types.ValueType.fromExternalString((String) val)); break;\n" } ]

[ { "change_type": "MODIFY", "old_path": "docs/site/builtins-reference.md", "new_path": "docs/site/builtins-reference.md", "diff": "@@ -56,6 +56,7 @@ limitations under the License.\n* [`lmCG`-Function](#lmcg-function)\n* [`lmDS`-Function](#lmds-function)\n* [`lmPredict`-Function](#lmPredict-function)\n+ * [`matrixProfile`-Function](#matrixProfile-function)\n* [`mdedup`-Function](#mdedup-function)\n* [`mice`-Function](#mice-function)\n* [`msvm`-Function](#msvm-function)\n@@ -197,7 +198,7 @@ y = toOneHot(X, numClasses)\n## `correctTypos`-Function\n-The `correctTypos` - function tries to correct typos in a given frame. This algorithm operates on the assumption that most strings are correct and simply swaps strings that do not occur often with similar strings that occur more often. If correct is set to FALSE only prints suggested corrections without effecting the frame.\n+The `correctTypos` - function tries to correct typos in a given frame. This algorithm operates on the assumption that most strings are correct and simply swaps strings that do not occur often with similar strings that occur more often. If correct is set to FALSE only prints suggested corrections without affecting the frame.\n### Usage\n@@ -1258,6 +1259,32 @@ yp = lmPredict(X = X, B = w, ytest=matrix(0,1,1))\n```\n+## `matrixProfile`-Function\n+\n+The `matrixProfile`-function implements the SCRIMP algorithm for efficient time-series analysis.\n+\n+### Usage\n+```r\n+matrixProfile(ts, window_size, sample_percent, is_verbose)\n+```\n+\n+### Arguments\n+| Name | Type | Default | Description |\n+| :------ | :------------- | -------- | :---------- |\n+| ts | Matrix | --- | Input Frame X |\n+| window_size | Integer | 4 | Sliding window size |\n+| sample_percent| Double | 1.0 | Degree of approximation between zero and one (1 computes the exact solution) |\n+| verbose | Boolean | False | Print debug information |\n+\n+### Returns\n+\n+| Type | Default | Description |\n+| :-------------- | -------- | :---------- |\n+| Matrix[Double] | --- | The computed matrix profile distances |\n+| Matrix[Integer] | --- | Indices of least distances |\n+\n+\n+\n## `mdedup`-Function\nThe `mdedup`-function implements builtin for deduplication using matching dependencies\n" }, { "change_type": "ADD", "old_path": null, "new_path": "scripts/builtin/matrixProfile.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+# ----------------------------------------------------------------------------\n+# References:\n+# Yan Zhu et al.. 2018.\n+# Matrix Profile XI: SCRIMP++: Time Series Motif Discovery at Interactive Speeds.\n+# 2018 IEEE International Conference on Data Mining (ICDM), 2018, pp. 837-846.\n+# DOI: 10.1109/ICDM.2018.00099.\n+# https://www.cs.ucr.edu/~eamonn/SCRIMP_ICDM_camera_ready_updated.pdf\n+# ----------------------------------------------------------------------------\n+\n+# Builtin function that computes the MatrixProfile of a time series efficiently\n+# using the SCRIMP++ algorithm.\n+#\n+# INPUT PARAMETERS:\n+# ----------------------------------------------------------------------------\n+# NAME TYPE DEFAULT MEANING\n+# ----------------------------------------------------------------------------\n+# ts Matrix --- Time series to profile\n+# window_size Integer 4 Sliding window size\n+# sample_percent Double 1.0 Degree of approximation\n+# between zero and one (1\n+# computes the exact solution)\n+# is_verbose Boolean False Print debug information\n+#\n+#\n+# RETURN VALUES\n+# ----------------------------------------------------------------------------\n+# NAME TYPE DEFAULT MEANING\n+# profile Matrix --- The computed matrix profile\n+# profile_index Matrix --- Indices of least distances\n+\n+\n+m_matrixProfile = function(Matrix[Double] ts, Integer window_size=4, Double sample_percent=1.0, Boolean is_verbose=FALSE)\n+ return(Matrix[Double] profile, Matrix[Double] profile_index)\n+{\n+ if (is_verbose)\n+ print (\"##############################\\n# MATRIXPROFILE SCRIPT ENTRY #\\n##############################\");\n+\n+ # TODO: preSCRIMP\n+ # requires a similarity search algorithm e.g.: MASS (Mueen's Algorithm for Similarity Search)\n+\n+ n = length(ts);\n+ [mu,sig] = moving_avg(ts, n, window_size);\n+ if (is_verbose) {\n+ print_ts(ts);\n+ print_ts(mu);\n+ print_ts(sig);\n+ }\n+\n+ # initialize\n+ profile_len = n-window_size+1;\n+ profile = matrix(Inf, cols=1, rows=profile_len);\n+ profile_index = matrix(1, cols=1, rows=profile_len);\n+\n+ # random permutation\n+ exclusion_zone = as.integer(ceil(window_size/4)) + 1;\n+ sample_size = profile_len-exclusion_zone;\n+ if (sample_percent < 1.0 & sample_percent >= 0.0) {\n+ sample_size = ceil(sample_size*sample_percent);\n+ }\n+ s = sample(sample_size, sample_size, FALSE);\n+ s = s + exclusion_zone;\n+\n+ if (is_verbose) {\n+ print(\"n: \" + n);\n+ print(\"window_size: \" + window_size);\n+ print(\"profile_len: \" + profile_len);\n+ print(\"exclusion_zone: \" + exclusion_zone);\n+ print(\"sample_size: \" + sample_size);\n+ }\n+ k_idx = 1;\n+ while (k_idx <= sample_size) {\n+ k = as.scalar(s[k_idx]);\n+ k_idx += 1;\n+ q = 0;\n+ for (i in 1:n-window_size+2-k) {\n+ if (i==1)\n+ q = as.scalar(t(ts[1:window_size]) %*% ts[k:k+window_size-1]);\n+ else\n+ q = as.scalar(q - ts[i-1]%*%ts[i+k-2] + ts[i+window_size-1]%*%ts[i+k+window_size-2]);\n+ d = sqrt(2*window_size*(1-(q - window_size*as.scalar(mu[i]*mu[i+k-1])) / (window_size*as.scalar(sig[i]*sig[i+k-1]))));\n+\n+ if (d < as.scalar(profile[i])) {\n+ profile[i] = d;\n+ profile_index[i] = as.matrix(i+k-1);\n+ }\n+ if (d < as.scalar(profile[i+k-1])) {\n+ profile[i+k-1] = d;\n+ profile_index[i+k-1] = i;\n+ }\n+ }\n+ }\n+\n+ print_ts(profile);\n+ print_ts(profile_index);\n+}\n+\n+moving_avg = function(Matrix[Double] array, Integer n, Integer window_size)\n+ return(Matrix[Double] mu, Matrix[Double] sig)\n+{\n+ profile_len = n - window_size + 1;\n+ cum_sum = matrix(0, cols=1, rows=n);\n+ sq_cum_sum = matrix(0, cols=1, rows=n);\n+ sums = matrix(0, cols=1, rows=profile_len);\n+ sq_sums = matrix(0, cols=1, rows=profile_len);\n+ mu = matrix(0, cols=1, rows=profile_len);\n+ sig_sq = matrix(0, cols=1, rows=profile_len);\n+ sig = matrix(0, cols=1, rows=profile_len);\n+\n+ cum_sum = cumsum(array);\n+ sq_cum_sum = cumsum(array*array);\n+\n+ sums[1] = cum_sum[window_size];\n+ sq_sums[1] = sq_cum_sum[window_size];\n+ for (i in 1:n-window_size) {\n+ sums[i+1] = cum_sum[window_size + i] - cum_sum[i];\n+ sq_sums[i+1] = sq_cum_sum[window_size + i] - sq_cum_sum[i];\n+ }\n+\n+ for (i in 1:profile_len) {\n+ mu[i] = sums[i] / window_size;\n+ sig_sq[i] = sq_sums[i] / window_size - mu[i] * mu[i];\n+ sig[i] = max(sqrt(sig_sq[i]), 0);\n+ }\n+}\n+\n+print_ts = function(Matrix[Double] ts) {\n+ print(toString(t(ts)));\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/common/Builtins.java", "new_path": "src/main/java/org/apache/sysds/common/Builtins.java", "diff": "@@ -177,6 +177,7 @@ public enum Builtins {\nLSTM_BACKWARD(\"lstm_backward\", false, ReturnType.MULTI_RETURN),\nLU(\"lu\", false, ReturnType.MULTI_RETURN),\nMAP(\"map\", false),\n+ MATRIXPROFILE(\"matrixProfile\", true),\nMAX(\"max\", \"pmax\", false),\nMAX_POOL(\"max_pool\", false),\nMAX_POOL_BACKWARD(\"max_pool_backward\", false),\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysds/test/functions/builtin/BuiltinMatrixProfileTest.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysds.test.functions.builtin;\n+\n+import org.apache.sysds.common.Types.ExecMode;\n+import org.apache.sysds.common.Types.ExecType;\n+import org.apache.sysds.runtime.matrix.data.MatrixValue.CellIndex;\n+import org.apache.sysds.test.AutomatedTestBase;\n+import org.apache.sysds.test.TestConfiguration;\n+import org.junit.Assert;\n+import org.junit.Test;\n+\n+import java.io.IOException;\n+import java.lang.Math;\n+import java.util.Random;\n+import java.util.Collections;\n+import java.util.Comparator;\n+import java.util.HashMap;\n+import java.util.LinkedList;\n+import java.util.List;\n+import java.util.Map.Entry;\n+\n+public class BuiltinMatrixProfileTest extends AutomatedTestBase\n+{\n+ private final static String TEST_NAME = \"matrix_profile\";\n+ private final static String TEST_DIR = \"functions/builtin/\";\n+ private static final String TEST_CLASS_DIR = TEST_DIR + BuiltinMatrixProfileTest.class.getSimpleName() + \"/\";\n+\n+ private static Random generator;\n+ private final static int seed = 42;\n+\n+ @Override\n+ public void setUp() {\n+ addTestConfiguration(TEST_NAME,new TestConfiguration(TEST_CLASS_DIR, TEST_NAME,new String[]{\"B\"}));\n+ }\n+\n+ @Test\n+ public void testMatrixProfileCP() throws IOException {\n+ runMatrixProfileTest(4, 1.0, \"TRUE\", ExecType.CP);\n+ }\n+\n+ @Test\n+ public void testMatrixProfileApproxCP() throws IOException {\n+ runMatrixProfileTest(4, 0.6, \"TRUE\", ExecType.CP);\n+ }\n+\n+// @Test\n+// public void testMatrixProfileSPARK() throws IOException {\n+// runMatrixProfileTest(4, 1.0, \"FALSE\", ExecType.SPARK);\n+// }\n+\n+\n+ private void runMatrixProfileTest(Integer window_size, Double sample_percent, String is_verbose, ExecType instType) throws IOException\n+ {\n+ ExecMode platformOld = setExecMode(instType);\n+\n+ try\n+ {\n+ loadTestConfiguration(getTestConfiguration(TEST_NAME));\n+\n+ String HOME = SCRIPT_DIR + TEST_DIR;\n+ fullDMLScriptName = HOME + TEST_NAME + \".dml\";\n+ programArgs = new String[]{\n+ \"-nvargs\", \"TS=\" + input(\"TS\"), \"MP=\" + output(\"MP\"),\n+ \"MPI=\" + output(\"MPI\"),\n+ \"window_size=\" + window_size,\n+ \"sample_percent=\" + sample_percent,\n+ \"is_verbose=\" + is_verbose};\n+\n+ generator = new Random(seed);\n+\n+ int len = 100;\n+ double[][] ts = genSineWave(len, 0.05, 1, 1);\n+ int[] noise_idxs = addNoise(1, len, ts);\n+ writeInputMatrixWithMTD(\"TS\", ts, false);\n+\n+ runTest(true, false, null, -1);\n+\n+ HashMap<CellIndex,Double> MP = readDMLMatrixFromOutputDir(\"MP\");\n+ @SuppressWarnings(\"unused\") //TODO\n+ HashMap<CellIndex,Double> MPI = readDMLMatrixFromOutputDir(\"MPI\");\n+\n+ List<Entry<CellIndex,Double>> sortedList = sortByValues(MP);\n+ Entry<CellIndex,Double> entry = sortedList.get(0);\n+ int highest_dist_idx = entry.getKey().row;\n+ int noise_idx = noise_idxs[0];\n+\n+ System.out.println(\"Detected anomaly around idx \" + highest_dist_idx);\n+ System.out.println(\"Noise idx: \" + noise_idx);\n+ Assert.assertTrue(highest_dist_idx>noise_idx-window_size && highest_dist_idx<noise_idx+window_size);\n+ }\n+ finally {\n+ rtplatform = platformOld;\n+ }\n+ }\n+\n+ private static double[][] genSineWave(Integer n, double sampling_rate, float p, float amp) {\n+ double[][] ts = new double[n][1];\n+ for (int i=0; i<n; ++i) {\n+ ts[i][0] = p*Math.sin(amp*sampling_rate*i);\n+ }\n+ return ts;\n+ }\n+\n+ private static int[] addNoise(Integer n, Integer len, double[][] ts) {\n+ int[] idxs = new int[n];\n+ for (int i=0; i<n; ++i) {\n+ int idx = generator.nextInt(len);\n+ ts[idx][0] += 1;\n+ idxs[i] = idx;\n+ }\n+ return idxs;\n+ }\n+\n+ private static List<Entry<CellIndex,Double>> sortByValues(HashMap<CellIndex,Double> map) {\n+ List<Entry<CellIndex,Double>> list = new LinkedList<>(map.entrySet());\n+ Collections.sort(list, new Comparator<Entry<CellIndex,Double>>() {\n+ public int compare(Entry<CellIndex,Double> o1, Entry<CellIndex,Double> o2) {\n+ return o2.getValue().compareTo(o1.getValue());\n+ }\n+ });\n+ return list;\n+ }\n+}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/builtin/matrix_profile.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+timeSeries = read($TS)\n+[MP,MPI] = matrixProfile(timeSeries, $window_size, $sample_percent, $is_verbose)\n+write(MP, $MP)\n+write(MPI, $MPI)\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/RandSPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/RandSPInstruction.java", "diff": "@@ -756,7 +756,8 @@ public class RandSPInstruction extends UnarySPInstruction {\nJavaPairRDD<MatrixIndexes, MatrixBlock> mbRDD =\nRDDConverterUtils.binaryCellToBinaryBlock(sec.getSparkContext(), miRDD, mcOut, true);\n- sec.getDataCharacteristics(output.getName()).setNonZeros(lrows);\n+ //step 5: output handling, incl meta data\n+ sec.getDataCharacteristics(output.getName()).set(mcOut);\nsec.setRDDHandleForVariable(output.getName(), mbRDD);\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/builtin/BuiltinMatrixProfileTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/builtin/BuiltinMatrixProfileTest.java", "diff": "@@ -61,10 +61,10 @@ public class BuiltinMatrixProfileTest extends AutomatedTestBase\nrunMatrixProfileTest(4, 0.6, \"TRUE\", ExecType.CP);\n}\n-// @Test\n-// public void testMatrixProfileSPARK() throws IOException {\n-// runMatrixProfileTest(4, 1.0, \"FALSE\", ExecType.SPARK);\n-// }\n+ @Test\n+ public void testMatrixProfileSPARK() throws IOException {\n+ runMatrixProfileTest(4, 0.02, \"FALSE\", ExecType.SPARK);\n+ }\nprivate void runMatrixProfileTest(Integer window_size, Double sample_percent, String is_verbose, ExecType instType) throws IOException\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/WriteSPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/WriteSPInstruction.java", "diff": "@@ -178,9 +178,9 @@ public class WriteSPInstruction extends SPInstruction implements LineageTraceabl\n//get input rdd\nJavaPairRDD<MatrixIndexes,MatrixBlock> in1 = sec.getBinaryMatrixBlockRDDHandleForVariable( input1.getName() );\nDataCharacteristics mc = sec.getDataCharacteristics(input1.getName());\n+ DataCharacteristics mcOut = mc; //by reference\n- if( fmt == FileFormat.MM || fmt == FileFormat.TEXT )\n- {\n+ if( fmt == FileFormat.MM || fmt == FileFormat.TEXT ) {\n//piggyback nnz maintenance on write\nLongAccumulator aNnz = null;\nif( !mc.nnzKnown() ) {\n@@ -208,16 +208,14 @@ public class WriteSPInstruction extends SPInstruction implements LineageTraceabl\nif( !mc.nnzKnown() )\nmc.setNonZeros( aNnz.value() );\n}\n- else if( fmt == FileFormat.CSV )\n- {\n+ else if( fmt == FileFormat.CSV ) {\nif( mc.getRows() == 0 || mc.getCols() == 0 ) {\nthrow new IOException(\"Write of matrices with zero rows or columns\"\n+ \" not supported (\"+mc.getRows()+\"x\"+mc.getCols()+\").\");\n}\n- LongAccumulator aNnz = null;\n-\n//piggyback nnz computation on actual write\n+ LongAccumulator aNnz = null;\nif( !mc.nnzKnown() ) {\naNnz = sec.getSparkContext().sc().longAccumulator(\"nnz\");\nin1 = in1.mapValues(new ComputeBinaryBlockNnzFunction(aNnz));\n@@ -234,9 +232,10 @@ public class WriteSPInstruction extends SPInstruction implements LineageTraceabl\nelse if( fmt == FileFormat.BINARY ) {\n//reblock output if needed\nint blen = Integer.parseInt(input4.getName());\n- DataCharacteristics mcOut = new MatrixCharacteristics(mc).setBlocksize(blen);\n- if( ConfigurationManager.getBlocksize() != blen )\n- in1 = RDDConverterUtils.binaryBlockToBinaryBlock(in1, mc, mcOut);\n+ boolean nonDefaultBlen = ConfigurationManager.getBlocksize() != blen;\n+ if( nonDefaultBlen )\n+ in1 = RDDConverterUtils.binaryBlockToBinaryBlock(in1, mc,\n+ new MatrixCharacteristics(mc).setBlocksize(blen));\n//piggyback nnz computation on actual write\nLongAccumulator aNnz = null;\n@@ -248,8 +247,10 @@ public class WriteSPInstruction extends SPInstruction implements LineageTraceabl\n//save binary block rdd on hdfs\nin1.saveAsHadoopFile(fname, MatrixIndexes.class, MatrixBlock.class, SequenceFileOutputFormat.class);\n- if(!mc.nnzKnown())\n+ if( !mc.nnzKnown() ) //update nnz\nmc.setNonZeros(aNnz.value().longValue());\n+ if( nonDefaultBlen )\n+ mcOut = new MatrixCharacteristics(mc).setBlocksize(blen);\n}\nelse if(fmt == FileFormat.LIBSVM) {\nif(mc.getRows() == 0 || mc.getCols() == 0) {\n@@ -258,9 +259,8 @@ public class WriteSPInstruction extends SPInstruction implements LineageTraceabl\n.getCols() + \").\");\n}\n- LongAccumulator aNnz = null;\n-\n//piggyback nnz computation on actual write\n+ LongAccumulator aNnz = null;\nif(!mc.nnzKnown()) {\naNnz = sec.getSparkContext().sc().longAccumulator(\"nnz\");\nin1 = in1.mapValues(new ComputeBinaryBlockNnzFunction(aNnz));\n@@ -280,7 +280,7 @@ public class WriteSPInstruction extends SPInstruction implements LineageTraceabl\n}\n// write meta data file\n- HDFSTool.writeMetaDataFile (fname + \".mtd\", ValueType.FP64, mc, fmt, formatProperties);\n+ HDFSTool.writeMetaDataFile(fname + \".mtd\", ValueType.FP64, mcOut, fmt, formatProperties);\n}\nprotected void processFrameWriteInstruction(SparkExecutionContext sec, String fname, FileFormat fmt, ValueType[] schema)\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/AutomatedTestBase.java", "new_path": "src/test/java/org/apache/sysds/test/AutomatedTestBase.java", "diff": "@@ -900,9 +900,15 @@ public abstract class AutomatedTestBase {\n}\npublic static void checkDMLMetaDataFile(String fileName, MatrixCharacteristics mc) {\n+ checkDMLMetaDataFile(fileName, mc, false);\n+ }\n+\n+ public static void checkDMLMetaDataFile(String fileName, MatrixCharacteristics mc, boolean checkBlocksize) {\nMatrixCharacteristics rmc = readDMLMetaDataFile(fileName);\nAssert.assertEquals(mc.getRows(), rmc.getRows());\nAssert.assertEquals(mc.getCols(), rmc.getCols());\n+ if( checkBlocksize )\n+ Assert.assertEquals(mc.getBlocksize(), rmc.getBlocksize());\n}\npublic static MatrixCharacteristics readDMLMetaDataFile(String fileName) {\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/io/binary/BlocksizeTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/io/binary/BlocksizeTest.java", "diff": "@@ -129,16 +129,16 @@ public class BlocksizeTest extends AutomatedTestBase\n//generate actual dataset\ndouble[][] X = getRandomMatrix(rows, cols, -1.0, 1.0, sparsity, 7);\nMatrixBlock mb = DataConverter.convertToMatrixBlock(X);\n- MatrixCharacteristics mc = new MatrixCharacteristics(rows, cols, inBlksize, inBlksize);\n+ MatrixCharacteristics mc = new MatrixCharacteristics(rows, cols, inBlksize);\nDataConverter.writeMatrixToHDFS(mb, input(\"X\"), FileFormat.BINARY, mc);\nHDFSTool.writeMetaDataFile(input(\"X.mtd\"), ValueType.FP64, mc, FileFormat.BINARY);\nrunTest(true, false, null, -1); //mult 7\n//compare matrices\n- checkDMLMetaDataFile(\"X\", new MatrixCharacteristics(rows, cols, outBlksize, outBlksize));\n+ checkDMLMetaDataFile(\"X\", new MatrixCharacteristics(rows, cols, outBlksize), true);\nMatrixBlock mb2 = DataConverter.readMatrixFromHDFS(\n- output(\"X\"), FileFormat.BINARY, rows, cols, outBlksize, outBlksize);\n+ output(\"X\"), FileFormat.BINARY, rows, cols, outBlksize, -1);\nfor( int i=0; i<mb.getNumRows(); i++ )\nfor( int j=0; j<mb.getNumColumns(); j++ ) {\ndouble val1 = mb.quickGetValue(i, j) * 7;\n" } ]

[ { "change_type": "MODIFY", "old_path": "docs/site/dml-language-reference.md", "new_path": "docs/site/dml-language-reference.md", "diff": "@@ -1638,7 +1638,7 @@ Scheme | Definition\n-------- | -----------\nDisjoint_Contiguous | For each worker, use a right indexing operation X[beg:end,] to obtain contiguous, non-overlapping partitions of rows\nDisjoint_Round_Robin | For each worker, use a permutation multiply or simpler a removeEmpty such as removeEmpty(target=X, margin=rows, select=(seq(1,nrow(X))%%k)==id)\n-Disjoint_Random | For each worker, use a permutation multiply P[beg:end,] %*% X, where P is constructed for example with P=table(seq(1,nrow(X),sample(nrow(X), nrow(X)))), i.e., sampling without replacement to ensure disjointness\n+Disjoint_Random | For each worker, use a permutation multiply P[beg:end,] %*% X, where P is constructed for example with P=table(seq(1,nrow(X)),sample(nrow(X), nrow(X))), i.e., sampling without replacement to ensure disjointness\nOverlap_Reshuffle | Similar to the above, except to create a new permutation matrix for each worker and without the indexing on P\n### Other Built-In Functions\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/QuaternaryFEDInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/QuaternaryFEDInstruction.java", "diff": "@@ -35,6 +35,8 @@ import org.apache.sysds.lops.WeightedSquaredLoss.WeightsType;\nimport org.apache.sysds.lops.WeightedUnaryMM;\nimport org.apache.sysds.lops.WeightedUnaryMM.WUMMType;\nimport org.apache.sysds.runtime.DMLRuntimeException;\n+import org.apache.sysds.runtime.controlprogram.caching.MatrixObject;\n+import org.apache.sysds.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysds.runtime.instructions.InstructionUtils;\nimport org.apache.sysds.runtime.instructions.cp.CPOperand;\nimport org.apache.sysds.runtime.matrix.operators.Operator;\n@@ -165,4 +167,12 @@ public abstract class QuaternaryFEDInstruction extends ComputationFEDInstruction\nreturn inst_str;\n}\n+\n+ protected void setOutputDataCharacteristics(MatrixObject X, MatrixObject U, MatrixObject V, ExecutionContext ec) {\n+ long rows = X.getNumRows() > 1 ? X.getNumRows() : U.getNumRows();\n+ long cols = X.getNumColumns() > 1 ? X.getNumColumns()\n+ : (U.getNumColumns() == V.getNumRows() ? V.getNumColumns() : V.getNumRows());\n+ MatrixObject out = ec.getMatrixObject(output);\n+ out.getDataCharacteristics().set(rows, cols, (int) X.getBlocksize());\n+ }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/QuaternaryWSigmoidFEDInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/QuaternaryWSigmoidFEDInstruction.java", "diff": "package org.apache.sysds.runtime.instructions.fed;\nimport java.util.ArrayList;\n-import java.util.concurrent.Future;\nimport org.apache.commons.lang3.ArrayUtils;\nimport org.apache.sysds.runtime.DMLRuntimeException;\nimport org.apache.sysds.runtime.controlprogram.caching.MatrixObject;\nimport org.apache.sysds.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysds.runtime.controlprogram.federated.FederatedRequest;\n-import org.apache.sysds.runtime.controlprogram.federated.FederatedRequest.RequestType;\n-import org.apache.sysds.runtime.controlprogram.federated.FederatedResponse;\nimport org.apache.sysds.runtime.controlprogram.federated.FederationMap;\nimport org.apache.sysds.runtime.controlprogram.federated.FederationMap.AlignType;\nimport org.apache.sysds.runtime.controlprogram.federated.FederationMap.FType;\n@@ -101,25 +98,24 @@ public class QuaternaryWSigmoidFEDInstruction extends QuaternaryFEDInstruction {\nFederatedRequest frComp = FederationUtils.callInstruction(instString,\noutput, new CPOperand[] {input1, input2, input3}, varNewIn);\n- // get partial results from federated workers\n- FederatedRequest frGet = new FederatedRequest(RequestType.GET_VAR, frComp.getID());\n-\nArrayList<FederatedRequest> frC = new ArrayList<>();\n- frC.add(fedMap.cleanup(getTID(), frComp.getID()));\nif(frSliced != null)\nfrC.add(fedMap.cleanup(getTID(), frSliced[0].getID()));\nfrC.add(fedMap.cleanup(getTID(), frB.getID()));\n- FederatedRequest[] frAll = ArrayUtils.addAll(new FederatedRequest[]{frB, frComp, frGet},\n+ FederatedRequest[] frAll = ArrayUtils.addAll(new FederatedRequest[]{frB, frComp},\nfrC.toArray(new FederatedRequest[0]));\n// execute federated instructions\n- Future<FederatedResponse>[] response = frSliced != null ?\n- fedMap.execute(getTID(), true, frSliced, frAll)\n- : fedMap.execute(getTID(), true, frAll);\n+ if(frSliced == null)\n+ fedMap.execute(getTID(), true, frAll);\n+ else\n+ fedMap.execute(getTID(), true, frSliced, frAll);\n- // bind partial results from federated responses\n- ec.setMatrixOutput(output.getName(), FederationUtils.bind(response, X.isFederated(FType.COL)));\n+ // derive output federated mapping\n+ MatrixObject out = ec.getMatrixObject(output);\n+ out.setFedMapping(fedMap.copyWithNewID(frComp.getID()));\n+ setOutputDataCharacteristics(X, U, V, ec);\n}\nelse {\nthrow new DMLRuntimeException(\"Unsupported federated inputs (X, U, V) = (\"\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/QuaternaryWUMMFEDInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/QuaternaryWUMMFEDInstruction.java", "diff": "package org.apache.sysds.runtime.instructions.fed;\nimport java.util.ArrayList;\n-import java.util.concurrent.Future;\nimport org.apache.commons.lang3.ArrayUtils;\nimport org.apache.sysds.runtime.DMLRuntimeException;\nimport org.apache.sysds.runtime.controlprogram.caching.MatrixObject;\nimport org.apache.sysds.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysds.runtime.controlprogram.federated.FederatedRequest;\n-import org.apache.sysds.runtime.controlprogram.federated.FederatedRequest.RequestType;\n-import org.apache.sysds.runtime.controlprogram.federated.FederatedResponse;\nimport org.apache.sysds.runtime.controlprogram.federated.FederationMap;\nimport org.apache.sysds.runtime.controlprogram.federated.FederationMap.AlignType;\nimport org.apache.sysds.runtime.controlprogram.federated.FederationMap.FType;\n@@ -72,6 +69,7 @@ public class QuaternaryWUMMFEDInstruction extends QuaternaryFEDInstruction {\nif(X.isFederated(FType.ROW)) { // row partitioned X\nif(U.isFederated(FType.ROW) && fedMap.isAligned(U.getFedMapping(), AlignType.ROW)) {\n+ // U federated and aligned\nvarNewIn[1] = U.getFedMapping().getID();\n}\nelse {\n@@ -85,6 +83,7 @@ public class QuaternaryWUMMFEDInstruction extends QuaternaryFEDInstruction {\nfrB = fedMap.broadcast(U);\nvarNewIn[1] = frB.getID();\nif(V.isFederated() && fedMap.isAligned(V.getFedMapping(), AlignType.COL, AlignType.COL_T)) {\n+ // V federated and aligned\nvarNewIn[2] = V.getFedMapping().getID();\n}\nelse {\n@@ -100,24 +99,24 @@ public class QuaternaryWUMMFEDInstruction extends QuaternaryFEDInstruction {\nFederatedRequest frComp = FederationUtils.callInstruction(instString, output,\nnew CPOperand[]{input1, input2, input3}, varNewIn);\n- // get partial results from federated workers\n- FederatedRequest frGet = new FederatedRequest(RequestType.GET_VAR, frComp.getID());\n-\nArrayList<FederatedRequest> frC = new ArrayList<>();\n- frC.add(fedMap.cleanup(getTID(), frComp.getID()));\nif(frSliced != null)\nfrC.add(fedMap.cleanup(getTID(), frSliced[0].getID()));\nfrC.add(fedMap.cleanup(getTID(), frB.getID()));\n- FederatedRequest[] frAll = ArrayUtils.addAll(new FederatedRequest[]{frB, frComp, frGet},\n+ FederatedRequest[] frAll = ArrayUtils.addAll(new FederatedRequest[]{frB, frComp},\nfrC.toArray(new FederatedRequest[0]));\n// execute federated instructions\n- Future<FederatedResponse>[] response = frSliced == null ?\n- fedMap.execute(getTID(), true, frAll) : fedMap.execute(getTID(), true, frSliced, frAll);\n+ if(frSliced == null)\n+ fedMap.execute(getTID(), true, frAll);\n+ else\n+ fedMap.execute(getTID(), true, frSliced, frAll);\n- // bind partial results from federated responses\n- ec.setMatrixOutput(output.getName(), FederationUtils.bind(response, X.isFederated(FType.COL)));\n+ // derive output federated mapping\n+ MatrixObject out = ec.getMatrixObject(output);\n+ out.setFedMapping(fedMap.copyWithNewID(frComp.getID()));\n+ setOutputDataCharacteristics(X, U, V, ec);\n}\nelse {\nthrow new DMLRuntimeException(\"Unsupported federated inputs (X, U, V) = (\"\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/federated/primitives/FederatedWeightedSigmoidTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/federated/primitives/FederatedWeightedSigmoidTest.java", "diff": "@@ -48,7 +48,7 @@ public class FederatedWeightedSigmoidTest extends AutomatedTestBase {\nprivate final static String OUTPUT_NAME = \"Z\";\n- private final static double TOLERANCE = 0;\n+ private final static double TOLERANCE = 1e-14;\nprivate final static int BLOCKSIZE = 1024;\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/federated/primitives/FederatedWeightedUnaryMatrixMultTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/federated/primitives/FederatedWeightedUnaryMatrixMultTest.java", "diff": "@@ -77,7 +77,7 @@ public class FederatedWeightedUnaryMatrixMultTest extends AutomatedTestBase\nreturn Arrays.asList(new Object[][] {\n// {rows, cols, rank, sparsity}\n{1202, 1003, 5, 0.001},\n- {1202, 1003, 5, 0.6}\n+ {1202, 1003, 5, 0.7}\n});\n}\n@@ -106,10 +106,11 @@ public class FederatedWeightedUnaryMatrixMultTest extends AutomatedTestBase\nfederatedWeightedUnaryMatrixMult(EXP_DIV_TEST_NAME, ExecMode.SPARK);\n}\n- @Test\n- public void federatedWeightedUnaryMatrixMultPow2SingleNode() {\n- federatedWeightedUnaryMatrixMult(POW_2_TEST_NAME, ExecMode.SINGLE_NODE);\n- }\n+ //TODO fix NaN issues in single node and spark\n+ // @Test\n+ // public void federatedWeightedUnaryMatrixMultPow2SingleNode() {\n+ // federatedWeightedUnaryMatrixMult(POW_2_TEST_NAME, ExecMode.SINGLE_NODE);\n+ // }\n// @Test\n// public void federatedWeightedUnaryMatrixMultPow2Spark() {\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/federated/quaternary/FederatedWUMMPow2Test.dml", "new_path": "src/test/scripts/functions/federated/quaternary/FederatedWUMMPow2Test.dml", "diff": "@@ -38,8 +38,6 @@ while(FALSE) { }\nZ3 = X / (V %*% t(U))^2;\nwhile(FALSE) { }\n-print(\"XX \"+mean(Z3))\nZ = Z1 + Z2 + mean(Z3);\n-print(\"XXX \"+as.scalar(Z[1,1]))\nwrite(Z, $out_Z);\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/federated/quaternary/FederatedWUMMPow2TestReference.dml", "new_path": "src/test/scripts/functions/federated/quaternary/FederatedWUMMPow2TestReference.dml", "diff": "@@ -33,9 +33,6 @@ X = t(X);\nZ3 = X / (V %*% t(U))^2;\n-print(\"XX \"+mean(Z3))\n-\nZ = Z1 + Z2 + mean(Z3);\n-print(\"XXX \"+as.scalar(Z[1,1]))\nwrite(Z, $out_Z);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/utils/RDDConverterUtils.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/utils/RDDConverterUtils.java", "diff": "@@ -188,8 +188,8 @@ public class RDDConverterUtils {\nJavaPairRDD<MatrixIndexes, MatrixBlock> in, DataCharacteristics mcIn, DataCharacteristics mcOut)\n{\nboolean shuffleFreeReblock = mcIn.dimsKnown() && mcOut.dimsKnown()\n- && (mcIn.getRows() < mcOut.getBlocksize() || mcIn.getBlocksize()%mcOut.getBlocksize() == 0)\n- && (mcIn.getCols() < mcOut.getBlocksize() || mcIn.getBlocksize()%mcOut.getBlocksize() == 0);\n+ && (mcIn.getRows() < mcIn.getBlocksize() || mcIn.getBlocksize()%mcOut.getBlocksize() == 0)\n+ && (mcIn.getCols() < mcIn.getBlocksize() || mcIn.getBlocksize()%mcOut.getBlocksize() == 0);\nJavaPairRDD<MatrixIndexes, MatrixBlock> out = in\n.flatMapToPair(new ExtractBlockForBinaryReblock(mcIn, mcOut));\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/io/binary/BlocksizeTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/io/binary/BlocksizeTest.java", "diff": "@@ -112,6 +112,21 @@ public class BlocksizeTest extends AutomatedTestBase\nrunBlocksizeTest(1000, 2000, ExecMode.SPARK);\n}\n+ @Test\n+ public void testSingleNode2xRowsCols() {\n+ runBlocksizeTest(1000, 7000, ExecMode.SINGLE_NODE);\n+ }\n+\n+ @Test\n+ public void testHybrid2xRowsCols() {\n+ runBlocksizeTest(1000, 7000, ExecMode.HYBRID);\n+ }\n+\n+ @Test\n+ public void testSpark2xRowsCols() {\n+ //test for invalid shuffle-free reblock\n+ runBlocksizeTest(1000, 7000, ExecMode.SPARK);\n+ }\nprivate void runBlocksizeTest(int inBlksize, int outBlksize, ExecMode mode)\n{\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/SPInstructionParser.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/SPInstructionParser.java", "diff": "@@ -59,6 +59,7 @@ import org.apache.sysds.runtime.instructions.spark.CpmmSPInstruction;\nimport org.apache.sysds.runtime.instructions.spark.CtableSPInstruction;\nimport org.apache.sysds.runtime.instructions.spark.CumulativeAggregateSPInstruction;\nimport org.apache.sysds.runtime.instructions.spark.CumulativeOffsetSPInstruction;\n+import org.apache.sysds.runtime.instructions.spark.DeCompressionSPInstruction;\nimport org.apache.sysds.runtime.instructions.spark.DnnSPInstruction;\nimport org.apache.sysds.runtime.instructions.spark.IndexingSPInstruction;\nimport org.apache.sysds.runtime.instructions.spark.MapmmChainSPInstruction;\n@@ -500,6 +501,9 @@ public class SPInstructionParser extends InstructionParser\ncase Compression:\nreturn CompressionSPInstruction.parseInstruction(str);\n+ case DeCompression:\n+ return DeCompressionSPInstruction.parseInstruction(str);\n+\ncase SpoofFused:\nreturn SpoofSPInstruction.parseInstruction(str);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/CompressionCPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/CompressionCPInstruction.java", "diff": "package org.apache.sysds.runtime.instructions.cp;\n+import org.apache.commons.lang3.tuple.Pair;\n+import org.apache.commons.logging.Log;\n+import org.apache.commons.logging.LogFactory;\nimport org.apache.sysds.hops.OptimizerUtils;\nimport org.apache.sysds.runtime.compress.CompressedMatrixBlockFactory;\n+import org.apache.sysds.runtime.compress.CompressionStatistics;\nimport org.apache.sysds.runtime.compress.SingletonLookupHashMap;\nimport org.apache.sysds.runtime.compress.workload.WTreeRoot;\nimport org.apache.sysds.runtime.controlprogram.context.ExecutionContext;\n@@ -29,9 +33,11 @@ import org.apache.sysds.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysds.runtime.matrix.operators.Operator;\npublic class CompressionCPInstruction extends ComputationCPInstruction {\n+ private static final Log LOG = LogFactory.getLog(CompressionCPInstruction.class.getName());\nprivate final int _singletonLookupID;\n+\nprivate CompressionCPInstruction(Operator op, CPOperand in, CPOperand out, String opcode, String istr,\nint singletonLookupID) {\nsuper(CPType.Compression, op, in, null, null, out, opcode, istr);\n@@ -61,8 +67,11 @@ public class CompressionCPInstruction extends ComputationCPInstruction {\nWTreeRoot root = (_singletonLookupID != 0) ? (WTreeRoot) m.get(_singletonLookupID) : null;\n// Compress the matrix block\n- MatrixBlock out = CompressedMatrixBlockFactory.compress(in, OptimizerUtils.getConstrainedNumThreads(-1), root)\n- .getLeft();\n+ Pair<MatrixBlock, CompressionStatistics> compResult = CompressedMatrixBlockFactory.compress(in, OptimizerUtils.getConstrainedNumThreads(-1), root);\n+\n+ if(LOG.isTraceEnabled())\n+ LOG.trace(compResult.getRight());\n+ MatrixBlock out = compResult.getLeft();\nm.removeKey(_singletonLookupID);\n// Set output and release input\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/CompressionSPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/CompressionSPInstruction.java", "diff": "package org.apache.sysds.runtime.instructions.spark;\n+import java.util.List;\n+\n+import org.apache.commons.logging.Log;\n+import org.apache.commons.logging.LogFactory;\nimport org.apache.spark.api.java.JavaPairRDD;\nimport org.apache.spark.api.java.function.Function;\nimport org.apache.sysds.runtime.compress.CompressedMatrixBlockFactory;\n@@ -35,7 +39,10 @@ import org.apache.sysds.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysds.runtime.matrix.data.MatrixIndexes;\nimport org.apache.sysds.runtime.matrix.operators.Operator;\n+import scala.Tuple2;\n+\npublic class CompressionSPInstruction extends UnarySPInstruction {\n+ private static final Log LOG = LogFactory.getLog(CompressionSPInstruction.class.getName());\nprivate final int _singletonLookupID;\n@@ -79,6 +86,11 @@ public class CompressionSPInstruction extends UnarySPInstruction {\n// execute compression\nJavaPairRDD<MatrixIndexes, MatrixBlock> out = in.mapValues(mappingFunction);\n+ if(LOG.isTraceEnabled()) {\n+ out.checkpoint();\n+ LOG.trace(\"\\nSpark compressed : \" + reduceSizes(out.mapValues(new SizeFunction()).collect())\n+ + \"\\nSpark uncompressed : \" + reduceSizes(in.mapValues(new SizeFunction()).collect()));\n+ }\n// set outputs\nsec.setRDDHandleForVariable(output.getName(), out);\n@@ -110,4 +122,26 @@ public class CompressionSPInstruction extends UnarySPInstruction {\n.getLeft();\n}\n}\n+\n+ public static class SizeFunction implements Function<MatrixBlock, Double> {\n+ private static final long serialVersionUID = 1L;\n+\n+ public SizeFunction() {\n+\n+ }\n+\n+ @Override\n+ public Double call(MatrixBlock arg0) throws Exception {\n+ return (double) arg0.getInMemorySize();\n+ }\n+ }\n+\n+ public static String reduceSizes(List<Tuple2<MatrixIndexes, Double>> in) {\n+ double sum = 0;\n+ for(Tuple2<MatrixIndexes, Double> e : in) {\n+ sum += e._2();\n+ }\n+\n+ return \"sum: \" + sum + \" mean: \" + (sum / in.size());\n+ }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/DeCompressionSPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/DeCompressionSPInstruction.java", "diff": "@@ -27,10 +27,10 @@ import org.apache.sysds.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysds.runtime.controlprogram.context.SparkExecutionContext;\nimport org.apache.sysds.runtime.instructions.InstructionUtils;\nimport org.apache.sysds.runtime.instructions.cp.CPOperand;\n-import org.apache.sysds.runtime.instructions.spark.CompressionSPInstruction.CompressionFunction;\nimport org.apache.sysds.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysds.runtime.matrix.data.MatrixIndexes;\nimport org.apache.sysds.runtime.matrix.operators.Operator;\n+import org.apache.sysds.utils.DMLCompressionStatistics;\npublic class DeCompressionSPInstruction extends UnarySPInstruction {\n@@ -51,9 +51,10 @@ public class DeCompressionSPInstruction extends UnarySPInstruction {\n// get input rdd handle\nJavaPairRDD<MatrixIndexes, MatrixBlock> in = sec.getBinaryMatrixBlockRDDHandleForVariable(input1.getName());\n- // execute compression\n- JavaPairRDD<MatrixIndexes, MatrixBlock> out = in.mapValues(new CompressionFunction());\n+ // execute decompression\n+ JavaPairRDD<MatrixIndexes, MatrixBlock> out = in.mapValues(new DeCompressionFunction());\n+ DMLCompressionStatistics.addDecompressSparkCount();\n// set outputs\nsec.setRDDHandleForVariable(output.getName(), out);\nsec.addLineageRDD(input1.getName(), output.getName());\n@@ -64,11 +65,10 @@ public class DeCompressionSPInstruction extends UnarySPInstruction {\n@Override\npublic MatrixBlock call(MatrixBlock arg0) throws Exception {\n- if(arg0 instanceof CompressedMatrixBlock){\n+ if(arg0 instanceof CompressedMatrixBlock)\nreturn ((CompressedMatrixBlock) arg0).decompress(OptimizerUtils.getConstrainedNumThreads(-1));\n- }else{\n+ else\nreturn arg0;\n}\n}\n}\n-}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/utils/DMLCompressionStatistics.java", "new_path": "src/main/java/org/apache/sysds/utils/DMLCompressionStatistics.java", "diff": "@@ -33,6 +33,9 @@ public class DMLCompressionStatistics {\nprivate static int DecompressMTCount = 0;\nprivate static double DecompressMT = 0.0;\n+ private static int DecompressSparkCount = 0;\n+ private static int DecompressCacheCount = 0;\n+\npublic static void reset() {\nPhase0 = 0.0;\nPhase1 = 0.0;\n@@ -44,6 +47,8 @@ public class DMLCompressionStatistics {\nDecompressST = 0.0;\nDecompressMTCount = 0;\nDecompressMT = 0.0;\n+ DecompressSparkCount = 0;\n+ DecompressCacheCount = 0;\n}\npublic static boolean haveCompressed(){\n@@ -85,12 +90,16 @@ public class DMLCompressionStatistics {\n}\n}\n- public static int getDecompressionCount() {\n- return DecompressMTCount;\n+ public static void addDecompressSparkCount(){\n+ DecompressSTCount++;\n}\n- public static int getDecompressionSTCount() {\n- return DecompressSTCount;\n+ public static void addDecompressCacheCount(){\n+ DecompressCacheCount++;\n+ }\n+\n+ public static int getDecompressionCount() {\n+ return DecompressMTCount + DecompressSTCount + DecompressSparkCount + DecompressCacheCount;\n}\npublic static void display(StringBuilder sb) {\n@@ -102,9 +111,11 @@ public class DMLCompressionStatistics {\nPhase3 / 1000,\nPhase4 / 1000,\nPhase5 / 1000));\n- sb.append(String.format(\"Decompression Counts (Single , Multi) thread :\\t%d/%d\\n\",\n+ sb.append(String.format(\"Decompression Counts (Single , Multi, Spark, Cache) thread :\\t%d/%d/%d/%d\\n\",\nDecompressSTCount,\n- DecompressMTCount));\n+ DecompressMTCount,\n+ DecompressSparkCount,\n+ DecompressCacheCount));\nsb.append(String.format(\"Dedicated Decompression Time (Single , Multi) thread :\\t%.3f/%.3f\\n\",\nDecompressST / 1000,\nDecompressMT / 1000));\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/compress/CompressInstructionRewrite.java", "new_path": "src/test/java/org/apache/sysds/test/functions/compress/CompressInstructionRewrite.java", "diff": "@@ -127,9 +127,7 @@ public class CompressInstructionRewrite extends AutomatedTestBase {\nif(LOG.isDebugEnabled())\nLOG.debug(stdout);\n- int decompressCount = 0;\n- decompressCount += DMLCompressionStatistics.getDecompressionCount();\n- decompressCount += DMLCompressionStatistics.getDecompressionSTCount();\n+ int decompressCount = DMLCompressionStatistics.getDecompressionCount();\nlong compressionCount = Statistics.getCPHeavyHitterCount(\"compress\");\nAssert.assertEquals(compressionCountsExpected, compressionCount);\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/compress/configuration/CompressBase.java", "new_path": "src/test/java/org/apache/sysds/test/functions/compress/configuration/CompressBase.java", "diff": "@@ -70,9 +70,7 @@ public abstract class CompressBase extends AutomatedTestBase {\nLOG.debug(runTest(null));\n- int decompressCount = 0;\n- decompressCount += DMLCompressionStatistics.getDecompressionCount();\n- decompressCount += DMLCompressionStatistics.getDecompressionSTCount();\n+ int decompressCount = DMLCompressionStatistics.getDecompressionCount();\nlong compressionCount = (instType == ExecType.SPARK) ? Statistics\n.getCPHeavyHitterCount(\"sp_compress\") : Statistics.getCPHeavyHitterCount(\"compress\");\nDMLCompressionStatistics.reset();\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/compress/workload/WorkloadAlgorithmTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/compress/workload/WorkloadAlgorithmTest.java", "diff": "@@ -55,11 +55,22 @@ public class WorkloadAlgorithmTest extends AutomatedTestBase {\nrunWorkloadAnalysisTest(TEST_NAME1, ExecMode.HYBRID, 2);\n}\n+\n+ @Test\n+ public void testLmSP() {\n+ runWorkloadAnalysisTest(TEST_NAME2, ExecMode.SPARK, 2);\n+ }\n+\n@Test\npublic void testLmCP() {\nrunWorkloadAnalysisTest(TEST_NAME2, ExecMode.HYBRID, 2);\n}\n+ @Test\n+ public void testPCASP() {\n+ runWorkloadAnalysisTest(TEST_NAME3, ExecMode.SPARK, 1);\n+ }\n+\n@Test\npublic void testPCACP() {\nrunWorkloadAnalysisTest(TEST_NAME3, ExecMode.HYBRID, 1);\n@@ -85,14 +96,15 @@ public class WorkloadAlgorithmTest extends AutomatedTestBase {\nwriteInputMatrixWithMTD(\"y\", y, false);\nString ret = runTest(null).toString();\n-\nif(ret.contains(\"ERROR:\"))\nfail(ret);\n// check various additional expectations\n- long actualCompressionCount = Statistics.getCPHeavyHitterCount(\"compress\");\n+ long actualCompressionCount = mode == ExecMode.HYBRID ? Statistics\n+ .getCPHeavyHitterCount(\"compress\") : Statistics.getCPHeavyHitterCount(\"sp_compress\");\n+\nAssert.assertEquals(compressionCount, actualCompressionCount);\n- Assert.assertTrue(heavyHittersContainsString(\"compress\"));\n+ Assert.assertTrue( mode == ExecMode.HYBRID ? heavyHittersContainsString(\"compress\") : heavyHittersContainsString(\"sp_compress\"));\nAssert.assertFalse(heavyHittersContainsString(\"m_scale\"));\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/compress/workload/SystemDS-config-compress-workload.xml", "new_path": "src/test/scripts/functions/compress/workload/SystemDS-config-compress-workload.xml", "diff": "<root>\n<sysds.compressed.linalg>workload</sysds.compressed.linalg>\n+ <sysds.defaultblocksize>8000</sysds.defaultblocksize>\n<sysds.cp.parallel.ops>true</sysds.cp.parallel.ops>\n<sysds.scratch>target/force_comp_scratch_space</sysds.scratch>\n</root>\n\\ No newline at end of file\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/compress/workload/WorkloadAnalysisMLogReg.dml", "new_path": "src/test/scripts/functions/compress/workload/WorkloadAnalysisMLogReg.dml", "diff": "@@ -27,7 +27,7 @@ print(\"\")\nprint(\"MLogReg\")\nX = scale(X=X, scale=TRUE, center=TRUE);\n-B = multiLogReg(X=X, Y=Y, verbose=TRUE, maxi = 10, maxii=10);\n+B = multiLogReg(X=X, Y=Y, verbose=FALSE, maxi=3, maxii=2);\n[nn, P, acc] = multiLogRegPredict(X=X, B=B, Y=Y)\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/builtin/bandit.dml", "new_path": "scripts/builtin/bandit.dml", "diff": "@@ -381,9 +381,9 @@ extractTopK = function(Frame[Unknown] pipeline, Matrix[Double] hyperparam,\n# remove the zero rows, identifiers of unique records\ndup = removeEmpty(target = dup, margin=\"rows\")\n# get the counts of duplicate tuples with their tuple id\n- countDist = table(dup, 1) > 0\n- countDist = countDist * seq(1, nrow(countDist))\n- countsVal = removeEmpty(target= countDist, margin=\"rows\")\n+ dist = table(dup, 1) > 0\n+ dist = dist * seq(1, nrow(dist))\n+ countsVal = removeEmpty(target= dist, margin=\"rows\")\nindexes = table(seq(1, nrow(countsVal)),countsVal,1,nrow(countsVal), cols=nrow(forDedup))\n# for each duplicate record just take the one reocrd and strip the others\n@@ -476,7 +476,8 @@ return (Matrix[Double] features)\n# OHE features\nOHE = sum(colMaxs(X) * mask)\nfeatures[1, 10] = OHE\n- distVal = countDistinct(Y)\n+ tab = table(Y, 1)\n+ distVal = nrow(tab)\nif(nrow(Y) > 1 & distVal <= 10)\n{\nctab = table(Y, 1)\n" }, { "change_type": "MODIFY", "old_path": "scripts/builtin/topk_cleaning.dml", "new_path": "scripts/builtin/topk_cleaning.dml", "diff": "@@ -121,8 +121,9 @@ s_topk_cleaning = function(Frame[Unknown] data, Frame[Unknown] metaData = as.fra\n\"4\", \"MVI\", \"OTLR\", \"MVI\", \"SCALE\"\n], rows=8, cols=5)\n-\n- if((nrow(Y) > 0 & countDistinct(Y) < 10))\n+ tab = table(Y, 1)\n+ dist = nrow(tab)\n+ if((nrow(Y) > 0 & dist < 10))\nlogical = logicalSeedCI\nelse\nlogical = logicalSeedNoCI\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkCleaningClassificationTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkCleaningClassificationTest.java", "diff": "@@ -47,7 +47,14 @@ public class BuiltinTopkCleaningClassificationTest extends AutomatedTestBase {\n@Test\npublic void testFindBestPipeline() {\n- runtopkCleaning(0.1, 3,5, TEST_NAME1, Types.ExecMode.SINGLE_NODE);\n+ runtopkCleaning(0.1, 3,5,\n+ TEST_NAME1, Types.ExecMode.SINGLE_NODE);\n+ }\n+\n+ @Ignore\n+ public void testFindBestPipelineHybrid() {\n+ runtopkCleaning(0.1, 3,5,\n+ TEST_NAME1, Types.ExecMode.HYBRID);\n}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkLogicalTest.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysds.test.functions.pipelines;\n+\n+import org.apache.sysds.common.Types;\n+import org.apache.sysds.test.AutomatedTestBase;\n+import org.apache.sysds.test.TestConfiguration;\n+import org.apache.sysds.test.TestUtils;\n+import org.junit.Assert;\n+import org.junit.Ignore;\n+import org.junit.Test;\n+\n+public class BuiltinTopkLogicalTest extends AutomatedTestBase {\n+ private final static String TEST_NAME = \"topkLogicalTest\";\n+ private final static String TEST_CLASS_DIR = SCRIPT_DIR + BuiltinTopkLogicalTest.class.getSimpleName() + \"/\";\n+\n+ private static final String RESOURCE = SCRIPT_DIR+\"functions/pipelines/\";\n+ private static final String DATA_DIR = DATASET_DIR+ \"pipelines/\";\n+\n+ private final static String DIRTY = DATA_DIR+ \"dirty.csv\";\n+ private final static String META = RESOURCE+ \"meta/meta_census.csv\";\n+\n+ private static final String PARAM_DIR = \"./scripts/pipelines/properties/\";\n+ private final static String PARAM = PARAM_DIR + \"param.csv\";\n+ private final static String PRIMITIVES = PARAM_DIR + \"testPrimitives.csv\";\n+ private final static String OUTPUT = RESOURCE+\"intermediates/logical.csv\";\n+\n+ private final static double dirtyScore = 0.7;\n+ @Override\n+ public void setUp() {\n+ addTestConfiguration(TEST_NAME,new TestConfiguration(TEST_CLASS_DIR, TEST_NAME,new String[]{\"R\"}));\n+ }\n+\n+ @Test\n+ public void testLogical1() {\n+ runTestLogical(10, 5, 2, Types.ExecMode.SINGLE_NODE);\n+ }\n+\n+ @Test\n+ public void testLogical2() {\n+ runTestLogical(2, 3, 2,\n+ Types.ExecMode.SINGLE_NODE);\n+ }\n+\n+ @Test\n+ public void testLogicalHybrid() {\n+ runTestLogical(3, 3, 2,\n+ Types.ExecMode.HYBRID);\n+ }\n+\n+ private void runTestLogical(int max_iter, int num_inst, int num_exec, Types.ExecMode et) {\n+\n+ // setOutputBuffering(true);\n+ String HOME = SCRIPT_DIR+\"functions/pipelines/\" ;\n+ Types.ExecMode modeOld = setExecMode(et);\n+ try {\n+ loadTestConfiguration(getTestConfiguration(TEST_NAME));\n+ fullDMLScriptName = HOME + TEST_NAME + \".dml\";\n+ programArgs = new String[] {\"-stats\", \"-exec\", \"singlenode\", \"-nvargs\", \"dirtyData=\"+DIRTY,\n+ \"metaData=\"+META, \"primitives=\"+PRIMITIVES, \"parameters=\"+PARAM, \"max_iter=\"+ max_iter,\n+ \"num_inst=\"+ num_inst, \"num_exec=\"+ num_exec,\n+ \"dirtyScore=\"+dirtyScore, \"output=\"+OUTPUT, \"O=\"+output(\"O\")};\n+\n+ runTest(true, EXCEPTION_NOT_EXPECTED, null, -1);\n+\n+ //expected loss smaller than default invocation\n+ Assert.assertTrue(TestUtils.readDMLBoolean(output(\"O\")));\n+ }\n+ finally {\n+ resetExecMode(modeOld);\n+ }\n+ }\n+}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/pipelines/topkLogicalTest.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+# Generate the logical pipelines for data cleaning\n+\n+source(\"scripts/pipelines/scripts/utils.dml\") as utils;\n+source(\"scripts/pipelines/scripts/enumerateLogical.dml\") as lg;\n+\n+\n+# read the inputs\n+X = read($dirtyData, data_type=\"frame\", format=\"csv\", header=TRUE,\n+ naStrings= [\"NA\", \"null\",\" \",\"NaN\", \"nan\", \"\", \"?\", \"99999\"]);\n+\n+metaInfo = read($metaData, data_type=\"frame\", format=\"csv\", header=FALSE);\n+primitives = read($primitives, data_type = \"frame\", format=\"csv\", header= TRUE)\n+param = read($parameters, data_type = \"frame\", format=\"csv\", header= TRUE)\n+dirtyScore = $dirtyScore\n+\n+max_iter = $max_iter\n+num_inst = $num_inst\n+num_exec = $num_exec\n+\n+\n+getSchema = metaInfo[1, 2:ncol(metaInfo)]\n+getMask = as.matrix(metaInfo[2, 2:ncol(metaInfo)])\n+getFdMask = as.matrix(metaInfo[3, 2:ncol(metaInfo)]) # columns of interest for FD computation\n+\n+# encode the categorical data\n+if(sum(getMask) > 0)\n+{\n+ # always recode the label\n+ index = vectorToCsv(getMask)\n+ jspecR = \"{ids:true, recode:[\"+index+\"]}\"\n+ [eX, X_meta] = transformencode(target=X, spec=jspecR);\n+ # change the schema to reflect the encoded values\n+ getSchema = map(getSchema, \"x->x.replace(\\\"STRING\\\", \\\"INT64\\\")\")\n+ getSchema = map(getSchema, \"x->x.replace(\\\"BOOLEAN\\\", \\\"INT64\\\")\")\n+\n+}\n+# if no categorical value exist then just cast the frame into matrix\n+else\n+ eX = as.matrix(X)\n+\n+# extract the class label\n+eY = eX[, ncol(eX)]\n+eX = eX[, 1:ncol(eX) - 1]\n+\n+print(\"y classes \\n\"+toString(table(eY, 1)))\n+getMask = getMask[, 1:ncol(getMask) - 1] # strip the mask of class label\n+getFdMask = getFdMask[, 1:ncol(getFdMask) - 1] # strip the mask of class label\n+getSchema = getSchema[, 1:ncol(getSchema) - 1] # strip the mask of class label\n+\n+metaList = list(mask=getMask, schema=getSchema, fd=as.matrix(0))\n+\n+logical = frame([\n+ \"1\", \"MVI\", \"0\", \"0\", \"0\", \"0\",\n+ # \"1\", \"OTLR\", \"0\", \"0\", \"0\", \"0\",\n+ # \"1\", \"CI\", \"0\", \"0\", \"0\", \"0\",\n+ # \"2\", \"MVI\", \"CI\", \"0\", \"0\", \"0\",\n+ # \"2\", \"MVI\", \"OTLR\", \"0\", \"0\", \"0\",\n+ # \"2\", \"MVI\", \"SCALE\", \"0\", \"0\", \"0\",\n+ # \"3\", \"MVI\", \"SCALE\", \"OTLR\", \"0\", \"0\",\n+ # \"4\", \"MVI\", \"OTLR\", \"CI\", \"SCALE\", \"0\",\n+ # \"4\", \"OTLR\", \"MVI\", \"CI\", \"SCALE\", \"0\",\n+ \"5\", \"MVI\", \"OTLR\", \"MVI\", \"CI\", \"SCALE\"\n+ ], rows=2, cols=6)\n+\n+\n+categories = frame([\"MVI\", \"OTLR\", \"SCALE\"], rows=1, cols=3)\n+cmr = matrix(\"4 0.7 1\", rows=1, cols=3)\n+[bestLogical, score, T] = lg::enumerateLogical(X=eX, y=eY, cmr=cmr, cat=categories, population=logical,\n+ max_iter=max_iter, metaList = metaList, evaluationFunc=\"evalClassification\", evalFunHp=matrix(\"1 1e-3 1e-9 100\", rows=1, cols=4),\n+ primitives=primitives, param=param , num_inst=num_inst, num_exec=num_exec, isTailed=TRUE, verbose=TRUE)\n+\n+print(\"score of pipeline: \"+toString(score)+\" in \"+(T/60000)+\" mins\")\n+print(\"bestLogical \"+toString(bestLogical))\n+\n+result = dirtyScore < score\n+print(\"result satisfied ------------\"+result)\n+\n+write(result , $O)\n+\n+\n+\n+# UDF for evaluation\n+# choice of parameters provided by API, X, Y, clone_X, evalFunHp (hyper-param), trainML (boolean for optimizing hp internally or passed by externally )\n+evalClassification = function(Matrix[Double] X, Matrix[Double] Y, Matrix[Double] Xorig, List[Unknown] metaList,\n+ Matrix[Double] evalFunHp, Integer trainML=0)\n+\n+return(Matrix[Double] output)\n+{\n+ cv = 2\n+ mask = as.matrix(metaList['mask'])\n+ print(\"min and max of y in eval: \"+min(Y)+\" \"+max(Y))\n+ if(max(Y) == min(Y)) {\n+ print(\"Y contains only one class\")\n+ accuracy = as.double(0)\n+ }\n+ else {\n+ if(trainML == 1)\n+ {\n+ # do the gridsearch for hyper-parameters\n+ params = list(\"icpt\", \"reg\", \"tol\", \"maxii\")\n+ paramRanges = list(seq(0, 2, 1), 10^seq(1,-4), 10^seq(1,-6), 10^seq(1,3));\n+\n+ trainArgs = list(X=X, Y=Y, icpt=-1, reg=-1, tol=-1, maxi=100, maxii=-1, verbose=FALSE);\n+ [B1, opt] = utils::topk_gridSearch(X=X, y=Y, train=\"multiLogReg\", predict=\"W\", numB=ncol(X)+1, cv=TRUE, cvk=cv,\n+ params=params, paramValues=paramRanges, trainArgs=trainArgs, verbose=FALSE);\n+ evalFunHp = as.matrix(opt)\n+ }\n+\n+ # do the k = 3 cross validations\n+ # evalFunHpM = as.matrix(evalFunHp)\n+ [accuracyMatrix] = crossV(X, Y, cv, evalFunHp, FALSE)\n+ accuracyMatrix = removeEmpty(target=accuracyMatrix, margin=\"rows\")\n+ score = mean(accuracyMatrix)\n+ print(cv +\" validation accuracy \"+score)\n+ }\n+ output = cbind(as.matrix(score), evalFunHp)\n+}\n+\n+# # ######################################################################\n+# # # # Function for cross validation using hold out method\n+# # # # Inputs: The input dataset X, Y and the value of k validation, mask of the\n+# # # # dataset for OHE of categorical columns, vector of ML hyper-parameters identified\n+# # # # via gridsearch and a boolean value of (un)weighted accuracy.\n+# # # # Output: It return a matrix having the accuracy of each fold.\n+# # ######################################################################\n+\n+crossV = function(Matrix[double] X, Matrix[double] y, Integer k, Matrix[Double] MLhp, Boolean isWeighted)\n+return (Matrix[Double] accuracyMatrix)\n+{\n+ accuracyMatrix = matrix(0, k, 1)\n+ dataList = list()\n+ testL = list()\n+ data = order(target = cbind(y, X), by = 1, decreasing=FALSE, index.return=FALSE)\n+ classes = table(data[, 1], 1)\n+ ins_per_fold = classes/k\n+ start_fold = matrix(1, rows=nrow(ins_per_fold), cols=1)\n+ fold_idxes = cbind(start_fold, ins_per_fold)\n+\n+ start_i = 0; end_i = 0; idx_fold = 1;;\n+ for(i in 1:k)\n+ {\n+ fold_i = matrix(0, 0, ncol(data))\n+ start=0; end=0;\n+ for(j in 1:nrow(classes))\n+ {\n+ idx = as.scalar(classes[j, 1])\n+ start = end + 1;\n+ end = end + idx\n+ class_j = data[start:end, ]\n+ start_i = as.scalar(fold_idxes[j, 1]);\n+ end_i = as.scalar(fold_idxes[j, 2])\n+ fold_i = rbind(fold_i, class_j[start_i:end_i, ])\n+ }\n+ dataList = append(dataList, fold_i)\n+ fold_idxes[, 1] = fold_idxes[, 2] + 1\n+ fold_idxes[, 2] += ins_per_fold\n+ }\n+\n+ for(i in seq(1,k))\n+ {\n+ [trainList, hold_out] = remove(dataList, i)\n+ trainset = rbind(trainList)\n+ testset = as.matrix(hold_out)\n+ trainX = trainset[, 2:ncol(trainset)]\n+ trainy = trainset[, 1]\n+ testX = testset[, 2:ncol(testset)]\n+ testy = testset[, 1]\n+ beta = multiLogReg(X=trainX, Y=trainy, icpt=as.scalar(MLhp[1,1]), reg=as.scalar(MLhp[1,2]), tol=as.scalar(MLhp[1,3]),\n+ maxi=as.scalar(MLhp[1,4]), maxii=50, verbose=FALSE);\n+ [prob, yhat, a] = multiLogRegPredict(testX, beta, testy, FALSE)\n+ accuracy = getAccuracy(testy, yhat, isWeighted)\n+ accuracyMatrix[i] = accuracy\n+ }\n+}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkCleaningClassificationTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkCleaningClassificationTest.java", "diff": "@@ -23,6 +23,7 @@ import org.apache.sysds.test.AutomatedTestBase;\nimport org.apache.sysds.test.TestConfiguration;\nimport org.apache.sysds.test.TestUtils;\nimport org.junit.Assert;\n+import org.junit.Ignore;\nimport org.junit.Test;\npublic class BuiltinTopkCleaningClassificationTest extends AutomatedTestBase {\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/builtin/bandit.dml", "new_path": "scripts/builtin/bandit.dml", "diff": "#\n#-------------------------------------------------------------\n-m_bandit = function(Matrix[Double] X_train, Matrix[Double] Y_train, List[Unknown] metaList, String evaluationFunc, Matrix[Double] evalFunHp,\n+m_bandit = function(Matrix[Double] X_train, Matrix[Double] Y_train, Matrix[Double] X_test, Matrix[Double] Y_test, List[Unknown] metaList, String evaluationFunc, Matrix[Double] evalFunHp,\nFrame[Unknown] lp, Frame[Unknown] primitives, Frame[Unknown] param, Integer k = 3, Integer R=50, Double baseLineScore,\nBoolean verbose = TRUE)\nreturn (Frame[Unknown] bestPipeline, Matrix[Double] bestHyperparams, Matrix[Double] bestAccuracy, Frame[String] feaFrameOuter)\n@@ -79,7 +79,7 @@ m_bandit = function(Matrix[Double] X_train, Matrix[Double] Y_train, List[Unknown\n}\nconfigurations = configurations[1:n_i, ]\n- [outPip,outHp, feaFrameOuter] = run_with_hyperparam(configurations, r_i, X_train, Y_train, metaList,\n+ [outPip,outHp, feaFrameOuter] = run_with_hyperparam(lp, configurations, r_i, X_train, Y_train, X_test, Y_test, metaList,\nevaluationFunc, evalFunHp, param, feaFrameOuter, verbose)\n# sort the pipelines by order of accuracy decreasing\na = order(target = outPip, by = 1, decreasing=TRUE, index.return=FALSE)\n@@ -195,7 +195,7 @@ get_physical_configurations = function(Frame[String] logical, Scalar[int] numCon\n}\n# this method will call the execute pipelines with their hyper-parameters\n-run_with_hyperparam = function(Frame[Unknown] ph_pip, Integer r_i, Matrix[Double] X, Matrix[Double] Y,\n+run_with_hyperparam = function(Frame[Unknown] lp, Frame[Unknown] ph_pip, Integer r_i, Matrix[Double] X, Matrix[Double] Y, Matrix[Double] Xtest, Matrix[Double] Ytest,\nList[Unknown] metaList, String evaluationFunc, Matrix[Double] evalFunHp, Frame[Unknown] param, Frame[Unknown] featureFrameOuter,\nBoolean verbose)\nreturn (Matrix[Double] output_operator, Matrix[Double] output_hyperparam, Frame[Unknown] featureFrameOuter) {\n@@ -207,6 +207,8 @@ run_with_hyperparam = function(Frame[Unknown] ph_pip, Integer r_i, Matrix[Double\n# rows in validation set\nclone_X = X\nclone_Y = Y\n+ clone_Xtest = Xtest\n+ clone_Ytest = Ytest\nindex = 1\nid = as.matrix(ph_pip[, 1])\nph_pip = ph_pip[, 2:ncol(ph_pip)]\n@@ -229,9 +231,13 @@ run_with_hyperparam = function(Frame[Unknown] ph_pip, Integer r_i, Matrix[Double\nindexes = cumsum(indexes)\nindexes = table(indexes, 1, 1, nrow(hp), 1)\nhp_matrix = removeEmpty(target = hp, margin=\"rows\", select = indexes)\n- [X, Y, T] = executePipeline(ph_pip[i], X, Y, as.matrix(metaList['mask']), as.matrix(metaList['fd']),\n- hp_matrix, no_of_flag_vars, FALSE)\n- argList = list(X=X, Y=Y, Xorig=clone_X, metaList=metaList, evalFunHp=evalFunHp, trainML=0)\n+ # # # clean the train data\n+ [X, Y, Tr] = executePipeline(lp, ph_pip[i], X, Y, as.matrix(metaList['mask']), as.matrix(metaList['fd']),\n+ hp_matrix, no_of_flag_vars, FALSE, FALSE)\n+ # # # clean the test data\n+ [Xtest, Ytest, T] = executePipeline(lp, ph_pip[i], Xtest, Ytest, as.matrix(metaList['mask']), as.matrix(metaList['fd']),\n+ hp_matrix, no_of_flag_vars, TRUE, FALSE)\n+ argList = list(X=X, Y=Y, Xtest=Xtest, Ytest=Ytest, Xorig=clone_X, metaList=metaList, evalFunHp=evalFunHp, trainML=0)\nt1 = time()\nevalFunOutput = eval(evaluationFunc, argList)\naccT = floor((time() - t1) / 1e+6)\n@@ -242,19 +248,23 @@ run_with_hyperparam = function(Frame[Unknown] ph_pip, Integer r_i, Matrix[Double\noutput_pipelines[index, ] = cbind(as.matrix(index), id[i,1])\nX = clone_X\nY = clone_Y\n+ Xtest = clone_Xtest\n+ Ytest = clone_Ytest\nindex = index + 1\nif(ncol(featureFrameOuter) > 1) {\nfeaFrame[r, 1:ncol(feaVec)] = as.frame(feaVec)\nfeaFrame[r, (ncol(feaVec)+1)] = pip_toString\nfeaFrame[r, (ncol(feaVec)+2)] = as.scalar(evalFunOutput[1, 1])\n- feaFrame[r, (ncol(feaVec)+3)] = T\n+ feaFrame[r, (ncol(feaVec)+3)] = Tr\nfeaFrame[r, (ncol(feaVec)+4)] = accT\n}\n}\nX = clone_X\nY = clone_Y\n+ Xtest = clone_Xtest\n+ Ytest = clone_Ytest\nif(ncol(featureFrameOuter) > 1)\nfeatureFrameOuter = rbind(featureFrameOuter, feaFrame)\n}\n" }, { "change_type": "MODIFY", "old_path": "scripts/builtin/executePipeline.dml", "new_path": "scripts/builtin/executePipeline.dml", "diff": "#\n#-------------------------------------------------------------\n-s_executePipeline = function(Frame[String] pipeline, Matrix[Double] X, Matrix[Double] Y, Matrix[Double] mask,\n- Matrix[Double] FD, Matrix[Double] hyperParameters, Integer flagsCount, Boolean verbose)\n+s_executePipeline = function(Frame[String] logical = as.frame(\"NULL\"), Frame[String] pipeline, Matrix[Double] X, Matrix[Double] Y, Matrix[Double] mask,\n+ Matrix[Double] FD, Matrix[Double] hyperParameters, Integer flagsCount, Boolean test = FALSE, Boolean verbose)\nreturn (Matrix[Double] X, Matrix[Double] Y, Double t2)\n{\nt1 = time();\n@@ -33,6 +33,8 @@ s_executePipeline = function(Frame[String] pipeline, Matrix[Double] X, Matrix[D\n}\nfor(i in 1:ncol(pipeline)) {\nop = as.scalar(pipeline[1,i])\n+ lgOp = as.scalar(logical[1,i])\n+ if(test == FALSE | lgOp != \"CI\") {\n[hp, withClass, dataFlag] = matrixToList(X, Y, mask, FD, hyperParameters[i], flagsCount, op)\nXclone = X\nX = eval(op, hp)\n@@ -46,6 +48,10 @@ s_executePipeline = function(Frame[String] pipeline, Matrix[Double] X, Matrix[D\nX = confirmMeta(X, mask)\n}\n+ else{\n+ print(\"not applying \"+lgOp+\" \"+op+\" on data test flag: \"+test)\n+ }\n+ }\nt2 = floor((time() - t1) / 1e+6)\nprint(\"PIPELINE EXECUTION ENDED: \"+t2+\" ms\")\n}\n" }, { "change_type": "MODIFY", "old_path": "scripts/builtin/topk_cleaning.dml", "new_path": "scripts/builtin/topk_cleaning.dml", "diff": "@@ -23,7 +23,7 @@ source(\"scripts/pipelines/scripts/utils.dml\") as utils;\nsource(\"scripts/pipelines/scripts/enumerateLogical.dml\") as lg;\n-s_topk_cleaning = function(Frame[Unknown] data, Frame[Unknown] metaData = as.frame(\"NULL\"), Frame[Unknown] primitives, Frame[Unknown] parameters,\n+s_topk_cleaning = function(Frame[Unknown] dataTrain, Frame[Unknown] dataTest, Frame[Unknown] metaData = as.frame(\"NULL\"), Frame[Unknown] primitives, Frame[Unknown] parameters,\nMatrix[Double] cmr = matrix(\"4 0.7 1\", rows=1, cols=3), String evaluationFunc, Matrix[Double] evalFunHp, Integer topK = 5,\nInteger resource_val = 20, Double sample = 0.1, Boolean isLastLabel = TRUE)\nreturn (Frame[Unknown] topKPipelines, Matrix[Double] topKHyperParams, Matrix[Double] topKScores, Frame[Unknown] bestLogical, Frame[Unknown] features, Double dirtyScore)\n@@ -33,8 +33,8 @@ s_topk_cleaning = function(Frame[Unknown] data, Frame[Unknown] metaData = as.fra\nif(as.scalar(metaData[1, 1]) == \"NULL\")\n{\nprint(\"creating meta data\")\n- r1 = detectSchema(data)\n- r2 = matrix(0, rows=1, cols=ncol(data))\n+ r1 = detectSchema(dataTrain)\n+ r2 = matrix(0, rows=1, cols=ncol(dataTrain))\nfor(i in 1 : ncol(r1))\n{\nif(as.scalar(r1[1, i]) == \"STRING\" | as.scalar(r1[1, i]) == \"BOOLEAN\")\n@@ -56,44 +56,62 @@ s_topk_cleaning = function(Frame[Unknown] data, Frame[Unknown] metaData = as.fra\n# separate the label\nif(isLastLabel) {\n- X = data[, 1:ncol(data) - 1]\n- label = data[, ncol(data)]\n+ Xtrain = dataTrain[, 1:ncol(dataTrain) - 1]\n+ labeltrain = dataTrain[, ncol(dataTrain)]\n+ Xtest = dataTest[, 1:ncol(dataTest) - 1]\n+ labeltest = dataTest[, ncol(dataTest)]\n# always recode the label\n- if(maskY == 1)\n- [Y, M] = transformencode(target=label, spec= \"{ids:true, recode:[1]}\");\n- else Y = as.matrix(label)\n+ if(maskY == 1) {\n+ [Ytrain, M] = transformencode(target=labeltrain, spec= \"{ids:true, recode:[1]}\");\n+ Ytest = transformapply(target=labeltest, spec= \"{ids:true, recode:[1]}\", meta=M);\n}\nelse\n{\n- X = data\n- Y = as.matrix(0)\n+ Ytrain = as.matrix(labeltrain)\n+ Ytest = as.matrix(labeltest)\n+ }\n+ }\n+ else\n+ {\n+ Xtrain = dataTrain\n+ Ytrain = as.matrix(0)\n+ Xtest = dataTest\n+ Ytest = as.matrix(0)\n}\n# # do the string processing\n- X_dirty = X\n- X = utils::stringProcessing(data=X, mask=mask, schema=schema, CorrectTypos=FALSE)\n+ X_train_dirty = Xtrain\n+ X_test_dirty = Xtest\n+ Xtrain = utils::stringProcessing(data=Xtrain, mask=mask, schema=schema, CorrectTypos=FALSE)\n+ Xtest = utils::stringProcessing(data=Xtest, mask=mask, schema=schema, CorrectTypos=FALSE)\n# # if mask has 1s then there are categorical features\nif(sum(mask) > 0)\n{\nindex = vectorToCsv(mask)\njspecR = \"{ids:true, recode:[\"+index+\"]}\"\n- [eX, X_meta] = transformencode(target=X, spec=jspecR);\n- [eX_dirty, X_meta_dirty] = transformencode(target=X_dirty, spec=jspecR);\n+ [eXtrain, X_meta] = transformencode(target=Xtrain, spec=jspecR);\n+ eXtest = transformapply(target=Xtest, spec=jspecR, meta=X_meta);\n+ [eX_train_dirty, X_meta_dirty] = transformencode(target=X_train_dirty, spec=jspecR);\n+ eX_test_dirty = transformapply(target=X_test_dirty, spec=jspecR, meta=X_meta_dirty);\n}\n# if no categorical value exist then just cast the frame into matrix\nelse {\n- eX = as.matrix(X)\n- eX_dirty = as.matrix(X_dirty)\n+ eXtrain = as.matrix(Xtrain)\n+ eX_train_dirty = as.matrix(X_train_dirty)\n+ eXtest = as.matrix(Xtest)\n+ eX_test_dirty = as.matrix(X_test_dirty)\n}\n# take the sample\n- [eX, Y] = utils::doSample(eX, Y, sample, TRUE)\n- [eX_dirty, Y] = utils::doSample(eX_dirty, Y, sample, FALSE)\n- eX_dirty = utils::dummycoding(eX_dirty, mask)\n+ [eXtrain, Ytrain] = utils::doSample(eXtrain, Ytrain, sample, TRUE)\n+ [eX_train_dirty, Ytrain] = utils::doSample(eX_train_dirty, Ytrain, sample, FALSE)\n+ # # allData = rbind(eX_train_dirty)\n+ # # eX_train_dirty = utils::dummycoding(eX_train_dirty, mask)\n+ # # eX_test_dirty = utils::dummycoding(eX_test_dirty, mask)\n# get the dirty score\n- scoreAndHp = eval(evaluationFunc, list(X=eX_dirty, Y=Y, Xorig=as.matrix(0), metaList=metaList, evalFunHp=evalFunHp, trainML=1))\n+ scoreAndHp = eval(evaluationFunc, list(X=eX_train_dirty, Y=Ytrain, Xtest=eX_test_dirty, Ytest=Ytest, Xorig=as.matrix(0), metaList=metaList, evalFunHp=evalFunHp, trainML=1))\ndirtyScore = as.scalar(scoreAndHp[1, 1])\nevalFunHp = scoreAndHp[1, 2:ncol(scoreAndHp)]\n@@ -121,9 +139,9 @@ s_topk_cleaning = function(Frame[Unknown] data, Frame[Unknown] metaData = as.fra\n\"4\", \"MVI\", \"OTLR\", \"MVI\", \"SCALE\"\n], rows=8, cols=5)\n- tab = table(Y, 1)\n+ tab = table(Ytrain, 1)\ndist = nrow(tab)\n- if((nrow(Y) > 0 & dist < 10))\n+ if((nrow(Ytrain) > 0 & dist < 10))\nlogical = logicalSeedCI\nelse\nlogical = logicalSeedNoCI\n@@ -132,15 +150,18 @@ s_topk_cleaning = function(Frame[Unknown] data, Frame[Unknown] metaData = as.fra\nidx = as.integer(as.scalar(logical[1, 1])) + 1\ncategory = logical[1, 2:idx]\n+ print(\"sending ytest in enumLog: \\n\"+toString(Ytest, rows=5))\n- [bestLogical, score, T] = lg::enumerateLogical(X=eX, y=Y, cmr=cmr, cat=category, population=logical,\n+ [bestLogical, score, T] = lg::enumerateLogical(X=eXtrain, y=Ytrain, Xtest=eXtest, ytest=Ytest, cmr=cmr, cat=category, population=logical,\nmax_iter=ceil(resource_val/topK), metaList = metaList, evaluationFunc=evaluationFunc, evalFunHp=evalFunHp,\n- primitives=primitives, param=parameters, num_inst = nrow(primitives), num_exec = topK, isTailed=TRUE, verbose=TRUE)\n+ primitives=primitives, param=parameters, num_inst = nrow(primitives), num_exec = topK, verbose=TRUE)\ntopKPipelines = as.frame(\"NULL\"); topKHyperParams = matrix(0,0,0); topKScores = matrix(0,0,0); features = as.frame(\"NULL\")\n- [topKPipelines, topKHyperParams, topKScores, features] = bandit(X_train=eX, Y_train=Y, metaList=metaList,\n+ [topKPipelines, topKHyperParams, topKScores, features] = bandit(X_train=eXtrain, Y_train=Ytrain, X_test=eXtest, Y_test=Ytest, metaList=metaList,\nevaluationFunc=evaluationFunc, evalFunHp=evalFunHp, lp=bestLogical, primitives=primitives, param=parameters, baseLineScore=dirtyScore,\nk=topK, R=resource_val, verbose=FALSE);\n}\n+\n+\n" }, { "change_type": "MODIFY", "old_path": "scripts/pipelines/scripts/enumerateLogical.dml", "new_path": "scripts/pipelines/scripts/enumerateLogical.dml", "diff": "source(\"scripts/builtin/bandit.dml\") as bandit;\n-enumerateLogical = function(Matrix[Double] X, Matrix[Double] y, Matrix[Double] cmr, Frame[Unknown] cat, Frame[Unknown] population,\n+enumerateLogical = function(Matrix[Double] X, Matrix[Double] y, Matrix[Double] Xtest, Matrix[Double] ytest, Matrix[Double] cmr, Frame[Unknown] cat, Frame[Unknown] population,\nInteger max_iter=10, List[Unknown] metaList, String evaluationFunc, Matrix[Double] evalFunHp, Frame[Unknown] primitives, Frame[Unknown] param,\n- Integer num_inst, Integer num_exec, Boolean isTailed = TRUE, Boolean verbose)\n+ Integer num_inst, Integer num_exec, Boolean verbose)\nreturn (Frame[Unknown] bestLg, Double pre_best, Double T)\n{\nt1 = time()\n@@ -66,6 +66,7 @@ return (Frame[Unknown] bestLg, Double pre_best, Double T)\nwhile(iter <= max_iter & !convergedOuter)\n{\nphysicalPipList = list()\n+ logicalPipList = list()\n# # # get the physical instances from logical ones\nfor(i in 1:nrow(population))\n{\n@@ -73,6 +74,7 @@ return (Frame[Unknown] bestLg, Double pre_best, Double T)\nlp = population[i, 2:lv]\nphysicalConf = bandit::get_physical_configurations(lp, num_inst, primitives)\nphysicalPipList = append(physicalPipList, physicalConf)\n+ logicalPipList = append(logicalPipList, lp)\n}\nscores = matrix(0, rows=length(physicalPipList), cols=1)\n@@ -81,15 +83,13 @@ return (Frame[Unknown] bestLg, Double pre_best, Double T)\nfor(i in 1:length(physicalPipList))\n{\nphysicalConf = as.frame(physicalPipList[i])\n+ lp = as.frame(logicalPipList[i])\n# # append configuration keys for extracting the pipeline later on\nid = seq(1, nrow(physicalConf))\nphysicalConf = cbind(as.frame(id), physicalConf)\n- if(isTailed) {\n- tail = frame([\"dummycoding\", \"m_pca\"], rows=nrow(physicalConf), cols=2)\n- physicalConf = cbind(physicalConf, tail)\n- }\n+\n# # execute the physical instances and store the minimum scores, each pipeline is executed num_exec times\n- [outPip,outHp, feaFrameOuter] = bandit::run_with_hyperparam(physicalConf, num_exec, X, y, metaList,\n+ [outPip,outHp, feaFrameOuter] = bandit::run_with_hyperparam(lp, physicalConf, num_exec, X, y, Xtest, ytest, metaList,\nevaluationFunc, evalFunHp, param, as.frame(\"\"), verbose)\n# # sort the configurations groupwise\nmax_perf = bandit::getMaxPerConf(outPip, nrow(physicalConf))\n@@ -104,11 +104,6 @@ return (Frame[Unknown] bestLg, Double pre_best, Double T)\nconvergedOuter = converged\nif(converged & (iter > 1))\n{\n- if(isTailed)\n- {\n- lg_tail = frame([\"DUMMY\", \"DIM\"], rows=1, cols=2)\n- bestLg = cbind(bestLg, lg_tail)\n- }\nprint(\"converged after \"+iter+\" iteration(s)\")\nprint(\"best score \" + pre_best)\nprint(\"best pipeline \" + toString(bestLg))\n@@ -151,11 +146,6 @@ return (Frame[Unknown] bestLg, Double pre_best, Double T)\n}\nif(pre_best == best_score) {\nprint(\"LogicalENumerator: did not converge after \"+max_iter+\" iterations\")\n- if(isTailed)\n- {\n- lg_tail = frame([\"DUMMY\", \"DIM\"], rows=1, cols=2)\n- bestLg = cbind(bestLg, lg_tail)\n- }\n}\n" }, { "change_type": "MODIFY", "old_path": "scripts/pipelines/scripts/utils.dml", "new_path": "scripts/pipelines/scripts/utils.dml", "diff": "@@ -267,7 +267,7 @@ return(Frame[Unknown] processedData)\n-topk_gridSearch = function(Matrix[Double] X, Matrix[Double] y, String train, String predict,\n+topk_gridSearch = function(Matrix[Double] X, Matrix[Double] y, Matrix[Double] Xtest=as.matrix(0), Matrix[Double] ytest=as.matrix(0), String train, String predict,\nInteger numB=ncol(X), List[String] params, List[Unknown] paramValues,\nList[Unknown] trainArgs = list(), List[Unknown] predictArgs = list(),\nBoolean cv = FALSE, Integer cvk = 5, Boolean verbose = TRUE)\n@@ -277,7 +277,7 @@ topk_gridSearch = function(Matrix[Double] X, Matrix[Double] y, String train, Str\nif( length(trainArgs) == 0 )\ntrainArgs = list(X=X, y=y, icpt=0, reg=-1, tol=-1, maxi=-1, verbose=FALSE);\nif( length(predictArgs) == 0 )\n- predictArgs = list(X, y);\n+ predictArgs = list(Xtest, ytest);\nif( cv & cvk <= 1 ) {\nprint(\"gridSearch: called with cv=TRUE but cvk=\"+cvk+\", set to default cvk=5.\")\ncvk = 5;\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkCleaningClassificationTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkCleaningClassificationTest.java", "diff": "@@ -23,11 +23,10 @@ import org.apache.sysds.test.AutomatedTestBase;\nimport org.apache.sysds.test.TestConfiguration;\nimport org.apache.sysds.test.TestUtils;\nimport org.junit.Assert;\n-import org.junit.Ignore;\nimport org.junit.Test;\npublic class BuiltinTopkCleaningClassificationTest extends AutomatedTestBase {\n- private final static String TEST_NAME1 = \"topkcleaningClassificationTest\";\n+ private final static String TEST_NAME = \"topkcleaningClassificationTest\";\nprivate final static String TEST_CLASS_DIR = SCRIPT_DIR + BuiltinTopkCleaningClassificationTest.class.getSimpleName() + \"/\";\nprivate final static String TEST_DIR = \"functions/pipelines/\";\n@@ -43,33 +42,39 @@ public class BuiltinTopkCleaningClassificationTest extends AutomatedTestBase {\n@Override\npublic void setUp() {\n- addTestConfiguration(TEST_NAME1,new TestConfiguration(TEST_CLASS_DIR, TEST_NAME1,new String[]{\"R\"}));\n+ addTestConfiguration(TEST_NAME, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME, new String[]{\"R\"}));\n}\n@Test\n- public void testFindBestPipeline() {\n+ public void testFindBestPipeline1() {\nruntopkCleaning(0.1, 3,5,\n- TEST_NAME1, Types.ExecMode.SINGLE_NODE);\n+ \"FALSE\", Types.ExecMode.SINGLE_NODE);\n}\n- @Ignore\n+ @Test\n+ public void testFindBestPipeline2() {\n+ runtopkCleaning(0.1, 3,5,\n+ \"TRUE\", Types.ExecMode.SINGLE_NODE);\n+ }\n+\n+ @Test\npublic void testFindBestPipelineHybrid() {\nruntopkCleaning(0.1, 3,5,\n- TEST_NAME1, Types.ExecMode.HYBRID);\n+ \"FALSE\", Types.ExecMode.HYBRID);\n}\n- private void runtopkCleaning(Double sample, int topk, int resources, String testName, Types.ExecMode et) {\n+ private void runtopkCleaning(Double sample, int topk, int resources, String cv, Types.ExecMode et) {\nsetOutputBuffering(true);\nTypes.ExecMode modeOld = setExecMode(et);\nString HOME = SCRIPT_DIR + TEST_DIR;\ntry {\n- loadTestConfiguration(getTestConfiguration(testName));\n- fullDMLScriptName = HOME + testName + \".dml\";\n+ loadTestConfiguration(getTestConfiguration(TEST_NAME));\n+ fullDMLScriptName = HOME + TEST_NAME + \".dml\";\nprogramArgs = new String[] {\"-stats\", \"-exec\", \"singlenode\", \"-nvargs\", \"dirtyData=\"+DIRTY,\n\"metaData=\"+META, \"primitives=\"+PRIMITIVES, \"parameters=\"+PARAM, \"topk=\"+ topk, \"rv=\"+ resources,\n- \"sample=\"+sample, \"O=\"+output(\"O\")};\n+ \"sample=\"+sample, \"testCV=\"+cv, \"O=\"+output(\"O\")};\nrunTest(true, EXCEPTION_NOT_EXPECTED, null, -1);\n@@ -81,5 +86,4 @@ public class BuiltinTopkCleaningClassificationTest extends AutomatedTestBase {\n}\n}\n-\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkCleaningRegressionTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkCleaningRegressionTest.java", "diff": "@@ -31,7 +31,6 @@ public class BuiltinTopkCleaningRegressionTest extends AutomatedTestBase{\nprivate final static String TEST_CLASS_DIR = SCRIPT_DIR + BuiltinTopkCleaningRegressionTest.class.getSimpleName() + \"/\";\nprivate static final String RESOURCE = SCRIPT_DIR+\"functions/pipelines/\";\n- // private static final String DATA_DIR = DATASET_DIR+ \"pipelines/\";\nprivate final static String DIRTY = DATASET_DIR+ \"Salaries.csv\";\nprivate final static String OUTPUT = RESOURCE+\"intermediates/\";\n@@ -50,7 +49,7 @@ public class BuiltinTopkCleaningRegressionTest extends AutomatedTestBase{\n\"lm\", Types.ExecMode.SINGLE_NODE);\n}\n- @Ignore\n+ @Test\npublic void testRegressionPipelinesHybrid() {\nrunFindPipelineTest(1.0, 5,5, 2,\n\"lm\", Types.ExecMode.HYBRID);\n@@ -60,7 +59,7 @@ public class BuiltinTopkCleaningRegressionTest extends AutomatedTestBase{\nprivate void runFindPipelineTest(Double sample, int topk, int resources, int crossfold,\nString target, Types.ExecMode et) {\n- // setOutputBuffering(true);\n+ setOutputBuffering(true);\nString HOME = SCRIPT_DIR+\"functions/pipelines/\" ;\nTypes.ExecMode modeOld = setExecMode(et);\ntry {\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkLogicalTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkLogicalTest.java", "diff": "@@ -66,7 +66,7 @@ public class BuiltinTopkLogicalTest extends AutomatedTestBase {\nprivate void runTestLogical(int max_iter, int num_inst, int num_exec, Types.ExecMode et) {\n- // setOutputBuffering(true);\n+ setOutputBuffering(true);\nString HOME = SCRIPT_DIR+\"functions/pipelines/\" ;\nTypes.ExecMode modeOld = setExecMode(et);\ntry {\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/topkLogicalTest.dml", "new_path": "src/test/scripts/functions/pipelines/topkLogicalTest.dml", "diff": "@@ -36,8 +36,7 @@ dirtyScore = $dirtyScore\nmax_iter = $max_iter\nnum_inst = $num_inst\nnum_exec = $num_exec\n-\n-\n+trainTestSplit = 0.7\ngetSchema = metaInfo[1, 2:ncol(metaInfo)]\ngetMask = as.matrix(metaInfo[2, 2:ncol(metaInfo)])\ngetFdMask = as.matrix(metaInfo[3, 2:ncol(metaInfo)]) # columns of interest for FD computation\n@@ -85,9 +84,18 @@ logical = frame([\ncategories = frame([\"MVI\", \"OTLR\", \"SCALE\"], rows=1, cols=3)\ncmr = matrix(\"4 0.7 1\", rows=1, cols=3)\n-[bestLogical, score, T] = lg::enumerateLogical(X=eX, y=eY, cmr=cmr, cat=categories, population=logical,\n+\n+# doing holdout evaluation\n+split = nrow(eX) * trainTestSplit\n+trainX = eX[1:split,]\n+trainY = eY[1:split,]\n+testX = eX[split+1:nrow(eX),]\n+testY = eY[split+1:nrow(eY),]\n+\n+\n+[bestLogical, score, T] = lg::enumerateLogical(X=trainX, y=trainY, Xtest=testX, ytest=testY, cmr=cmr, cat=categories, population=logical,\nmax_iter=max_iter, metaList = metaList, evaluationFunc=\"evalClassification\", evalFunHp=matrix(\"1 1e-3 1e-9 100\", rows=1, cols=4),\n- primitives=primitives, param=param , num_inst=num_inst, num_exec=num_exec, isTailed=TRUE, verbose=TRUE)\n+ primitives=primitives, param=param , num_inst=num_inst, num_exec=num_exec, verbose=TRUE)\nprint(\"score of pipeline: \"+toString(score)+\" in \"+(T/60000)+\" mins\")\nprint(\"bestLogical \"+toString(bestLogical))\n@@ -101,7 +109,7 @@ write(result , $O)\n# UDF for evaluation\n# choice of parameters provided by API, X, Y, clone_X, evalFunHp (hyper-param), trainML (boolean for optimizing hp internally or passed by externally )\n-evalClassification = function(Matrix[Double] X, Matrix[Double] Y, Matrix[Double] Xorig, List[Unknown] metaList,\n+evalClassification = function(Matrix[Double] X, Matrix[Double] Y, Matrix[Double] Xtest, Matrix[Double] Ytest, Matrix[Double] Xorig, List[Unknown] metaList,\nMatrix[Double] evalFunHp, Integer trainML=0)\nreturn(Matrix[Double] output)\n@@ -121,74 +129,19 @@ return(Matrix[Double] output)\nparamRanges = list(seq(0, 2, 1), 10^seq(1,-4), 10^seq(1,-6), 10^seq(1,3));\ntrainArgs = list(X=X, Y=Y, icpt=-1, reg=-1, tol=-1, maxi=100, maxii=-1, verbose=FALSE);\n- [B1, opt] = utils::topk_gridSearch(X=X, y=Y, train=\"multiLogReg\", predict=\"W\", numB=ncol(X)+1, cv=TRUE, cvk=cv,\n+ [B1, opt] = utils::topk_gridSearch(X=X, y=Y, Xtest=Xtest, ytest=Ytest, train=\"multiLogReg\", predict=\"accuracy\", numB=ncol(X)+1, cv=FALSE, cvk=cv,\nparams=params, paramValues=paramRanges, trainArgs=trainArgs, verbose=FALSE);\nevalFunHp = as.matrix(opt)\n}\n- # do the k = 3 cross validations\n- # evalFunHpM = as.matrix(evalFunHp)\n- [accuracyMatrix] = crossV(X, Y, cv, evalFunHp, FALSE)\n- accuracyMatrix = removeEmpty(target=accuracyMatrix, margin=\"rows\")\n- score = mean(accuracyMatrix)\n- print(cv +\" validation accuracy \"+score)\n- }\n- output = cbind(as.matrix(score), evalFunHp)\n-}\n-\n-# # ######################################################################\n-# # # # Function for cross validation using hold out method\n-# # # # Inputs: The input dataset X, Y and the value of k validation, mask of the\n-# # # # dataset for OHE of categorical columns, vector of ML hyper-parameters identified\n-# # # # via gridsearch and a boolean value of (un)weighted accuracy.\n-# # # # Output: It return a matrix having the accuracy of each fold.\n-# # ######################################################################\n+ beta = multiLogReg(X=X, Y=Y, icpt=as.scalar(evalFunHp[1,1]), reg=as.scalar(evalFunHp[1,2]), tol=as.scalar(evalFunHp[1,3]),\n+ maxi=as.scalar(evalFunHp[1,4]), maxii=50, verbose=FALSE);\n+ [prob, yhat, score] = multiLogRegPredict(Xtest, beta, Ytest, FALSE)\n-crossV = function(Matrix[double] X, Matrix[double] y, Integer k, Matrix[Double] MLhp, Boolean isWeighted)\n-return (Matrix[Double] accuracyMatrix)\n-{\n- accuracyMatrix = matrix(0, k, 1)\n- dataList = list()\n- testL = list()\n- data = order(target = cbind(y, X), by = 1, decreasing=FALSE, index.return=FALSE)\n- classes = table(data[, 1], 1)\n- ins_per_fold = classes/k\n- start_fold = matrix(1, rows=nrow(ins_per_fold), cols=1)\n- fold_idxes = cbind(start_fold, ins_per_fold)\n-\n- start_i = 0; end_i = 0; idx_fold = 1;;\n- for(i in 1:k)\n- {\n- fold_i = matrix(0, 0, ncol(data))\n- start=0; end=0;\n- for(j in 1:nrow(classes))\n- {\n- idx = as.scalar(classes[j, 1])\n- start = end + 1;\n- end = end + idx\n- class_j = data[start:end, ]\n- start_i = as.scalar(fold_idxes[j, 1]);\n- end_i = as.scalar(fold_idxes[j, 2])\n- fold_i = rbind(fold_i, class_j[start_i:end_i, ])\n}\n- dataList = append(dataList, fold_i)\n- fold_idxes[, 1] = fold_idxes[, 2] + 1\n- fold_idxes[, 2] += ins_per_fold\n- }\n-\n- for(i in seq(1,k))\n- {\n- [trainList, hold_out] = remove(dataList, i)\n- trainset = rbind(trainList)\n- testset = as.matrix(hold_out)\n- trainX = trainset[, 2:ncol(trainset)]\n- trainy = trainset[, 1]\n- testX = testset[, 2:ncol(testset)]\n- testy = testset[, 1]\n- beta = multiLogReg(X=trainX, Y=trainy, icpt=as.scalar(MLhp[1,1]), reg=as.scalar(MLhp[1,2]), tol=as.scalar(MLhp[1,3]),\n- maxi=as.scalar(MLhp[1,4]), maxii=50, verbose=FALSE);\n- [prob, yhat, a] = multiLogRegPredict(testX, beta, testy, FALSE)\n- accuracy = getAccuracy(testy, yhat, isWeighted)\n- accuracyMatrix[i] = accuracy\n+ output = cbind(as.matrix(score), evalFunHp)\n}\n+accuracy = function(Matrix[Double] X, Matrix[Double] y, Matrix[Double] B) return (Matrix[Double] err) {\n+ [M,yhat,acc] = multiLogRegPredict(X=X, B=B, Y=y, verbose=TRUE);\n+ err = as.matrix(1-(acc/100));\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/topkcleaningClassificationTest.dml", "new_path": "src/test/scripts/functions/pipelines/topkcleaningClassificationTest.dml", "diff": "@@ -32,14 +32,27 @@ param = read($parameters, data_type = \"frame\", format=\"csv\", header= TRUE)\ntopK = $topk\nresources = $rv\nsample=$sample\n+testCV = as.logical($testCV)\n+trainTestSplit = 0.7\n+\n+if(testCV)\n+ evalFunc = \"evalClassificationCV\"\n+\n+else\n+ evalFunc = \"evalClassification\"\n+\n+split = nrow(F) * trainTestSplit\n+trainData = F[1:split,]\n+testData = F[split+1:nrow(F),]\nif(nrow(metaInfo) < 2)\nstop(\"incomplete meta info\")\nmetaInfo = metaInfo[, 2:ncol(metaInfo)]\n+# # # split in train/test 70/30\n-[topKPipelines, topKHyperParams, topKScores, bestLogical, features, dirtyScore] = topk_cleaning(F, metaInfo, primitives, param,\n- matrix(\"2 0.7 1\", rows=1, cols=3), \"evalClassification\", as.matrix(\"0\"), topK, resources, sample, TRUE)\n+[topKPipelines, topKHyperParams, topKScores, bestLogical, features, dirtyScore] = topk_cleaning(trainData, testData, metaInfo, primitives, param,\n+ matrix(\"2 0.7 1\", rows=1, cols=3), evalFunc, as.matrix(\"0\"), topK, resources, sample, TRUE)\nprint(\"dirty accuracy \"+toString(dirtyScore))\n@@ -48,7 +61,7 @@ print(\"topk pipelines \"+toString(topKPipelines))\nprint(\"topk hyper params \"+toString(topKHyperParams))\nprint(\"topk scores: \\n\"+toString(topKScores))\nperf = as.double(as.scalar(topKScores[1, 1])) - as.double(dirtyScore)\n-print(\"performce improvemnet \"+ perf)\n+print(\"performance improvement \"+ perf)\nresult = dirtyScore < as.scalar(topKScores[1, 1])\nwrite(result, $O)\n@@ -56,14 +69,16 @@ write(result, $O)\n# UDF for evaluation\n# choice of parameters provided by API, X, Y, clone_X, evalFunHp (hyper-param), trainML (boolean for optimizing hp internally or passed by externally )\n-evalClassification = function(Matrix[Double] X, Matrix[Double] Y, Matrix[Double] Xorig, List[Unknown] metaList,\n+evalClassificationCV = function(Matrix[Double] X, Matrix[Double] Y, Matrix[Double] Xtest, Matrix[Double] Ytest, Matrix[Double] Xorig, List[Unknown] metaList,\nMatrix[Double] evalFunHp, Integer trainML=0)\nreturn(Matrix[Double] output)\n{\ncv = 2\nmask = as.matrix(metaList['mask'])\n- print(\"min and max of y in eval: \"+min(Y)+\" \"+max(Y))\n+ X = utils::dummycoding(replace(target = rbind(X, Xtest), pattern = NaN, replacement=1), mask)\n+ Y = rbind(Y, Ytest)\n+\nif(max(Y) == min(Y)) {\nprint(\"Y contains only one class\")\naccuracy = as.double(0)\n@@ -74,9 +89,6 @@ return(Matrix[Double] output)\n# do the gridsearch for hyper-parameters\nparams = list(\"icpt\", \"reg\", \"tol\", \"maxii\")\nparamRanges = list(seq(0, 2, 1), 10^seq(1,-4), 10^seq(1,-6), 10^seq(1,3));\n-\n- # if(sum(mask) > 0)\n- # X = utils::dummycoding(replace(target = X, pattern = NaN, replacement=0), mask)\ntrainArgs = list(X=X, Y=Y, icpt=-1, reg=-1, tol=-1, maxi=100, maxii=-1, verbose=FALSE);\n[B1, opt] = utils::topk_gridSearch(X=X, y=Y, train=\"multiLogReg\", predict=\"W\", numB=ncol(X)+1, cv=TRUE, cvk=cv,\nparams=params, paramValues=paramRanges, trainArgs=trainArgs, verbose=FALSE);\n@@ -149,7 +161,61 @@ return (Matrix[Double] accuracyMatrix)\naccuracy = getAccuracy(testy, yhat, isWeighted)\naccuracyMatrix[i] = accuracy\n}\n+}\n+\n+# UDF for evaluation\n+# choice of parameters provided by API, X, Y, clone_X, evalFunHp (hyper-param), trainML (boolean for optimizing hp internally or passed by externally )\n+evalClassification = function(Matrix[Double] X, Matrix[Double] Y, Matrix[Double] Xtest, Matrix[Double] Ytest, Matrix[Double] Xorig, List[Unknown] metaList,\n+ Matrix[Double] evalFunHp, Integer trainML=0)\n+\n+return(Matrix[Double] output)\n+{\n+ mask = as.matrix(metaList['mask'])\n+ if(sum(mask) > 0)\n+ {\n+ X = replace(target=X, pattern=NaN, replacement=1)\n+ Xtest = replace(target=Xtest, pattern=NaN, replacement=1)\n+ idx = vectorToCsv(mask)\n+ # specifications for one-hot encoding of categorical features\n+ jspecDC = \"{ids:true, dummycode:[\"+idx+\"]}\";\n+ # OHE of categorical features\n+ [dX, dM] = transformencode(target=as.frame(rbind(X,Xtest)), spec=jspecDC);\n+ X = dX[1:nrow(X),]\n+ Xtest = dX[nrow(X)+1:nrow(dX),]\n+ }\n+\n+ print(\"cols in X and Xtest: \"+ncol(X)+\" \"+ncol(Xtest))\n+ if(ncol(X) != ncol(Xtest))\n+ stop(\"Dimension mismatch: number of columns and train and test are not equal\")\n+ cv = 2\n+\n+ print(\"min and max of y in eval: \"+min(Y)+\" \"+max(Y))\n+ if(max(Y) == min(Y)) {\n+ print(\"Y contains only one class\")\n+ accuracy = as.double(0)\n+ }\n+ else {\n+ if(trainML == 1)\n+ {\n+ # do the gridsearch for hyper-parameters\n+ params = list(\"icpt\", \"reg\", \"tol\", \"maxii\")\n+ paramRanges = list(seq(0, 2, 1), 10^seq(1,-4), 10^seq(1,-6), 10^seq(1,3));\n+ trainArgs = list(X=rbind(X, Xtest), y=rbind(Y, Ytest), Xtest=Xtest, ytest=Ytest, icpt=-1, reg=-1, tol=-1, maxi=100, maxii=-1, verbose=FALSE);\n+ [B1, opt] = utils::topk_gridSearch(X=X, y=Y, train=\"multiLogReg\", predict=\"W\", numB=ncol(X)+1, cv=TRUE, cvk=cv,\n+ params=params, paramValues=paramRanges, trainArgs=trainArgs, verbose=FALSE);\n+ evalFunHp = as.matrix(opt)\n+ }\n+\n+ # do the hold out train/test\n+ # evalFunHpM = as.matrix(evalFunHp)\n+ beta = multiLogReg(X=X, Y=Y, icpt=as.scalar(evalFunHp[1,1]), reg=as.scalar(evalFunHp[1,2]), tol=as.scalar(evalFunHp[1,3]),\n+ maxi=as.scalar(evalFunHp[1,4]), maxii=50, verbose=FALSE);\n+\n+ [prob, yhat, score] = multiLogRegPredict(Xtest, beta, Ytest, FALSE)\n}\n+ output = cbind(as.matrix(score), evalFunHp)\n+ print(\"hold out accuracy: \"+score)\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/topkcleaningRegressionTest.dml", "new_path": "src/test/scripts/functions/pipelines/topkcleaningRegressionTest.dml", "diff": "@@ -32,7 +32,11 @@ topK = $topk\nresources = $rv\nsample=$sample\n-[topKPipelines, topKHyperParams, topKScores, bestLogical, features, dirtyScore] = topk_cleaning(data=F, primitives=primitives, parameters=param,\n+split = nrow(F) * 0.7\n+trainData = F[1:split,]\n+testData = F[split+1:nrow(F),]\n+\n+[topKPipelines, topKHyperParams, topKScores, bestLogical, features, dirtyScore] = topk_cleaning(dataTrain=trainData, dataTest=testData, primitives=primitives, parameters=param,\ncmr=matrix(\"4 0.7 1\", rows=1, cols=3), evaluationFunc=\"evalRegression\", evalFunHp=as.matrix(\"0\"), topK=topK, resource_val=resources, sample=sample, isLastLabel=TRUE)\n@@ -50,7 +54,7 @@ write(result, $O)\n# UDF for evaluation\n# choice of parameters provided by API, X, Y, clone_X, evalFunHp (hyper-param), trainML (boolean for optimizing hp internally or passed by externally )\n-evalRegression = function(Matrix[Double] X, Matrix[Double] Y, Matrix[Double] Xorig, List[Unknown] metaList,\n+evalRegression = function(Matrix[Double] X, Matrix[Double] Y, Matrix[Double] Xtest, Matrix[Double] Ytest, Matrix[Double] Xorig, List[Unknown] metaList,\nMatrix[Double] evalFunHp, Integer trainML=0)\nreturn(Matrix[Double] output)\n@@ -58,6 +62,9 @@ return(Matrix[Double] output)\ncv = 2\nmask = as.matrix(metaList['mask'])\n+ X = utils::dummycoding(replace(target = rbind(X, Xtest), pattern = NaN, replacement=1), mask)\n+ Y = rbind(Y, Ytest)\n+\nif(max(Y) == min(Y)) {\nprint(\"Y contains only one class\")\naccuracy = as.double(0)\n@@ -69,7 +76,7 @@ return(Matrix[Double] output)\nparams = list(\"icpt\",\"reg\", \"tol\", \"maxi\");\nparamRanges = list(seq(0,2),10^seq(0,-4), 10^seq(-6,-12), 10^seq(1,3));\n[B1, opt] = utils::topk_gridSearch(X=X, y=Y, train=\"lm\", predict=\"wmape\",\n- numB=ncol(X)+1, cv=TRUE, params=params, paramValues=paramRanges);\n+ numB=ncol(X)+1, cv=TRUE, params=params, paramValues=paramRanges, verbose=FALSE);\nevalFunHp = as.matrix(opt)\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/ipa/InterProceduralAnalysis.java", "new_path": "src/main/java/org/apache/sysds/hops/ipa/InterProceduralAnalysis.java", "diff": "@@ -241,7 +241,8 @@ public class InterProceduralAnalysis\nFunctionCallGraph graph2 = new FunctionCallGraph(_prog);\nList<IPAPass> fpasses = Arrays.asList(\nnew IPAPassRemoveUnusedFunctions(),\n- new IPAPassCompressionWorkloadAnalysis());\n+ new IPAPassCompressionWorkloadAnalysis(), // workload-aware compression\n+ new IPAPassApplyStaticAndDynamicHopRewrites()); //split after compress\nfor(IPAPass pass : fpasses)\nif( pass.isApplicable(graph2) )\npass.rewriteProgram(_prog, graph2, null);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/rewrite/RewriteSplitDagDataDependentOperators.java", "new_path": "src/main/java/org/apache/sysds/hops/rewrite/RewriteSplitDagDataDependentOperators.java", "diff": "@@ -34,6 +34,7 @@ import org.apache.sysds.common.Types.OpOpN;\nimport org.apache.sysds.common.Types.ParamBuiltinOp;\nimport org.apache.sysds.common.Types.ReOrgOp;\nimport org.apache.sysds.conf.ConfigurationManager;\n+import org.apache.sysds.conf.DMLConfig;\nimport org.apache.sysds.hops.AggBinaryOp;\nimport org.apache.sysds.hops.DataOp;\nimport org.apache.sysds.hops.Hop;\n@@ -42,6 +43,7 @@ import org.apache.sysds.hops.LiteralOp;\nimport org.apache.sysds.hops.ParameterizedBuiltinOp;\nimport org.apache.sysds.hops.TernaryOp;\nimport org.apache.sysds.hops.recompile.Recompiler;\n+import org.apache.sysds.lops.Compression.CompressConfig;\nimport org.apache.sysds.parser.DataIdentifier;\nimport org.apache.sysds.parser.StatementBlock;\nimport org.apache.sysds.parser.VariableSet;\n@@ -75,7 +77,10 @@ public class RewriteSplitDagDataDependentOperators extends StatementBlockRewrite\npublic List<StatementBlock> rewriteStatementBlock(StatementBlock sb, ProgramRewriteStatus state)\n{\n//DAG splits not required for forced single node\n- if( DMLScript.getGlobalExecMode() == ExecMode.SINGLE_NODE\n+ CompressConfig compress = CompressConfig.valueOf(ConfigurationManager\n+ .getDMLConfig().getTextValue(DMLConfig.COMPRESSED_LINALG).toUpperCase());\n+ if( (DMLScript.getGlobalExecMode() == ExecMode.SINGLE_NODE\n+ && !(compress != CompressConfig.FALSE) )\n|| !HopRewriteUtils.isLastLevelStatementBlock(sb) )\nreturn Arrays.asList(sb);\n@@ -225,7 +230,8 @@ public class RewriteSplitDagDataDependentOperators extends StatementBlockRewrite\nreturn;\n//prevent unnecessary dag split (dims known or no consumer operations)\n- boolean noSplitRequired = ( hop.dimsKnown() || HopRewriteUtils.hasOnlyWriteParents(hop, true, true) );\n+ boolean noSplitRequired = (HopRewriteUtils.hasOnlyWriteParents(hop, true, true)\n+ || hop.dimsKnown() || DMLScript.getGlobalExecMode() == ExecMode.SINGLE_NODE);\nboolean investigateChilds = true;\n//collect data dependent operations (to be extended as necessary)\n@@ -294,14 +300,8 @@ public class RewriteSplitDagDataDependentOperators extends StatementBlockRewrite\n}\n}\n- //#4 second-order eval function\n- if( HopRewriteUtils.isNary(hop, OpOpN.EVAL) && !noSplitRequired ) {\n- cand.add(hop);\n- investigateChilds = false;\n- }\n-\n- //#5 sql\n- if( hop instanceof DataOp && ((DataOp) hop).getOp() == OpOpData.SQLREAD && !noSplitRequired) {\n+ //#4 other data dependent operators (default handling)\n+ if( isBasicDataDependentOperator(hop, noSplitRequired) ) {\ncand.add(hop);\ninvestigateChilds = false;\n}\n@@ -315,6 +315,14 @@ public class RewriteSplitDagDataDependentOperators extends StatementBlockRewrite\nhop.setVisited();\n}\n+ private static boolean isBasicDataDependentOperator(Hop hop, boolean noSplitRequired) {\n+ return (HopRewriteUtils.isNary(hop, OpOpN.EVAL) & !noSplitRequired)\n+ || (HopRewriteUtils.isData(hop, OpOpData.SQLREAD) & !noSplitRequired)\n+ || (hop.requiresCompression() & !HopRewriteUtils.hasOnlyWriteParents(hop, true, true));\n+ //note: for compression we probe for write parents (part of noSplitRequired) directly\n+ // because we want to split even if the dimensions are known\n+ }\n+\nprivate static boolean hasTransientWriteParents( Hop hop ) {\nfor( Hop p : hop.getParent() )\nif( p instanceof DataOp && ((DataOp)p).getOp()==OpOpData.TRANSIENTWRITE )\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/compress/workload/WorkloadAlgorithmTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/compress/workload/WorkloadAlgorithmTest.java", "diff": "@@ -33,8 +33,6 @@ import org.junit.Test;\npublic class WorkloadAlgorithmTest extends AutomatedTestBase {\n- // private static final Log LOG = LogFactory.getLog(WorkloadAnalysisTest.class.getName());\n-\nprivate final static String TEST_NAME1 = \"WorkloadAnalysisMLogReg\";\nprivate final static String TEST_NAME2 = \"WorkloadAnalysisLm\";\nprivate final static String TEST_NAME3 = \"WorkloadAnalysisPCA\";\n@@ -55,7 +53,6 @@ public class WorkloadAlgorithmTest extends AutomatedTestBase {\nrunWorkloadAnalysisTest(TEST_NAME1, ExecMode.HYBRID, 2);\n}\n-\n@Test\npublic void testLmSP() {\nrunWorkloadAnalysisTest(TEST_NAME2, ExecMode.SPARK, 2);\n@@ -80,12 +77,12 @@ public class WorkloadAlgorithmTest extends AutomatedTestBase {\nExecMode oldPlatform = setExecMode(mode);\ntry {\n-\nloadTestConfiguration(getTestConfiguration(testname));\nString HOME = SCRIPT_DIR + TEST_DIR;\nfullDMLScriptName = HOME + testname + \".dml\";\n- programArgs = new String[] {\"-stats\", \"20\", \"-args\", input(\"X\"), input(\"y\"), output(\"B\")};\n+ programArgs = new String[] {\"-explain\",\"-stats\",\n+ \"20\", \"-args\", input(\"X\"), input(\"y\"), output(\"B\")};\ndouble[][] X = TestUtils.round(getRandomMatrix(10000, 20, 0, 10, 1.0, 7));\nwriteInputMatrixWithMTD(\"X\", X, false);\n@@ -95,9 +92,7 @@ public class WorkloadAlgorithmTest extends AutomatedTestBase {\n}\nwriteInputMatrixWithMTD(\"y\", y, false);\n- String ret = runTest(null).toString();\n- if(ret.contains(\"ERROR:\"))\n- fail(ret);\n+ runTest(null);\n// check various additional expectations\nlong actualCompressionCount = mode == ExecMode.HYBRID ? Statistics\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/StringInitCPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/StringInitCPInstruction.java", "diff": "@@ -21,11 +21,16 @@ package org.apache.sysds.runtime.instructions.cp;\nimport java.util.StringTokenizer;\n+import org.apache.commons.lang3.tuple.Pair;\n+import org.apache.sysds.common.Types.DataType;\n+import org.apache.sysds.common.Types.ValueType;\nimport org.apache.sysds.lops.DataGen;\nimport org.apache.sysds.lops.Lop;\nimport org.apache.sysds.runtime.DMLRuntimeException;\nimport org.apache.sysds.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysds.runtime.instructions.InstructionUtils;\n+import org.apache.sysds.runtime.lineage.LineageItem;\n+import org.apache.sysds.runtime.lineage.LineageItemUtils;\nimport org.apache.sysds.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysds.runtime.matrix.operators.Operator;\n@@ -97,4 +102,14 @@ public class StringInitCPInstruction extends UnaryCPInstruction {\n//put output into symbol table\nec.setMatrixOutput(outName, outBlk);\n}\n+\n+ @Override\n+ public Pair<String, LineageItem> getLineageItem(ExecutionContext ec) {\n+ CPOperand rlen = new CPOperand(String.valueOf(getRows()), ValueType.INT64, DataType.SCALAR, true);\n+ CPOperand clen = new CPOperand(String.valueOf(getCols()), ValueType.INT64, DataType.SCALAR, true);\n+ CPOperand data = new CPOperand(_data, ValueType.STRING, DataType.SCALAR, true);\n+ return Pair.of(output.getName(),\n+ new LineageItem(getOpcode(), LineageItemUtils.getLineage(ec, rlen, clen, data)));\n+\n+ }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/lineage/LineageRecomputeUtils.java", "new_path": "src/main/java/org/apache/sysds/runtime/lineage/LineageRecomputeUtils.java", "diff": "@@ -403,6 +403,16 @@ public class LineageRecomputeUtils {\noperands.put(item.getId(), operands.get(item.getInputs()[0].getId()));\nbreak;\n}\n+ case StringInit: {\n+ HashMap<String, Hop> params = new HashMap<>();\n+ params.put(DataExpression.RAND_ROWS, operands.get(item.getInputs()[0].getId()));\n+ params.put(DataExpression.RAND_COLS, operands.get(item.getInputs()[1].getId()));\n+ params.put(DataExpression.RAND_MIN, operands.get(item.getInputs()[2].getId()));\n+ params.put(DataExpression.RAND_MAX, operands.get(item.getInputs()[2].getId()));\n+ Hop datagen = new DataGenOp(OpOpDG.SINIT, new DataIdentifier(\"tmp\"), params);\n+ operands.put(item.getId(), datagen);\n+ break;\n+ }\ndefault:\nthrow new DMLRuntimeException(\"Unsupported instruction \"\n+ \"type: \" + ctype.name() + \" (\" + item.getOpcode() + \").\");\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/lineage/LineageTraceExecTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/lineage/LineageTraceExecTest.java", "diff": "@@ -43,6 +43,7 @@ public class LineageTraceExecTest extends LineageBase {\nprotected static final String TEST_NAME4 = \"LineageTraceExec4\"; //rand - matrix result - unspecified seed\nprotected static final String TEST_NAME5 = \"LineageTraceExec5\"; //rand - scalar result - unspecified seed\nprotected static final String TEST_NAME6 = \"LineageTraceExec6\"; //nary rbind\n+ protected static final String TEST_NAME7 = \"LineageTraceExec7\"; //sinit\nprotected String TEST_CLASS_DIR = TEST_DIR + LineageTraceExecTest.class.getSimpleName() + \"/\";\n@@ -62,6 +63,7 @@ public class LineageTraceExecTest extends LineageBase {\naddTestConfiguration( TEST_NAME4, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME4, new String[] {\"R\"}) );\naddTestConfiguration( TEST_NAME5, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME5, new String[] {\"R\"}) );\naddTestConfiguration( TEST_NAME6, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME6, new String[] {\"R\"}) );\n+ addTestConfiguration( TEST_NAME7, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME7, new String[] {\"R\"}) );\n}\n@Test\n@@ -94,6 +96,11 @@ public class LineageTraceExecTest extends LineageBase {\ntestLineageTraceExec(TEST_NAME6);\n}\n+ @Test\n+ public void testLineageTraceExec7() {\n+ testLineageTraceExec(TEST_NAME7);\n+ }\n+\nprivate void testLineageTraceExec(String testname) {\nLOG.debug(\"------------ BEGIN \" + testname + \"------------\");\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/lineage/LineageTraceExec7.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+X = matrix(\"1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4\", rows=4, cols=4);\n+R = X + 1;\n+write(R, $2);\n+\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/paramserv/FederatedPSControlThread.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/paramserv/FederatedPSControlThread.java", "diff": "@@ -321,7 +321,7 @@ public class FederatedPSControlThread extends PSWorker implements Callable<Void>\nreturn _ps.pull(_workerID);\n}\n- protected void weighAndPushGradients(ListObject gradients) {\n+ protected void weightAndPushGradients(ListObject gradients) {\n// scale gradients - must only include MatrixObjects\nif(_weighting && _weightingFactor != 1) {\nTiming tWeighting = DMLScript.STATISTICS ? new Timing(true) : null;\n@@ -354,7 +354,7 @@ public class FederatedPSControlThread extends PSWorker implements Callable<Void>\nint localStartBatchNum = getNextLocalBatchNum(currentLocalBatchNumber++, _possibleBatchesPerLocalEpoch);\nListObject model = pullModel();\nListObject gradients = computeGradientsForNBatches(model, 1, localStartBatchNum);\n- weighAndPushGradients(gradients);\n+ weightAndPushGradients(gradients);\nParamservUtils.cleanupListObject(model);\nParamservUtils.cleanupListObject(gradients);\n}\n@@ -378,7 +378,7 @@ public class FederatedPSControlThread extends PSWorker implements Callable<Void>\n// Pull the global parameters from ps\nListObject model = pullModel();\nListObject gradients = computeGradientsForNBatches(model, _numBatchesPerEpoch, localStartBatchNum, true);\n- weighAndPushGradients(gradients);\n+ weightAndPushGradients(gradients);\nParamservUtils.cleanupListObject(model);\nParamservUtils.cleanupListObject(gradients);\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/federated/paramserv/FederatedParamservTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/federated/paramserv/FederatedParamservTest.java", "diff": "@@ -64,7 +64,6 @@ public class FederatedParamservTest extends AutomatedTestBase {\nreturn Arrays.asList(new Object[][] {\n// Network type, number of federated workers, data set size, batch size, epochs, learning rate, update type, update frequency\n// basic functionality\n- //{\"TwoNN\", 4, 60000, 32, 4, 0.01, \"BSP\", \"BATCH\", \"KEEP_DATA_ON_WORKER\", \"NONE\" , \"false\",\"BALANCED\", 200},\n{\"TwoNN\", 2, 4, 1, 4, 0.01, \"BSP\", \"BATCH\", \"KEEP_DATA_ON_WORKER\", \"BASELINE\", \"true\", \"IMBALANCED\", 200},\n{\"CNN\", 2, 4, 1, 4, 0.01, \"BSP\", \"EPOCH\", \"SHUFFLE\", \"NONE\", \"true\", \"IMBALANCED\", 200},\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/builtin/tomeklink.dml", "new_path": "scripts/builtin/tomeklink.dml", "diff": "# NAME TYPE DEFAULT MEANING\n# ---------------------------------------------------------------------------------------------\n# X MATRIX --- Data Matrix (nxm)\n-# y MATRIX --- Label Matrix (nx1)\n+# y MATRIX --- Label Matrix (nx1), greater than zero\n# ---------------------------------------------------------------------------------------------\n# OUTPUT:\n# X_under - Data Matrix without Tomek links\n# drop_idx - Indices of dropped rows/labels wrt input\n-###### MAIN PART ######\n-\nm_tomeklink = function(Matrix[Double] X, Matrix[Double] y)\nreturn (Matrix[Double] X_under, Matrix[Double] y_under, Matrix[Double] drop_idx) {\n- majority_label = 0\n- n = nrow(X)\n- m = ncol(X)\n- tomek_links = get_links(X, y, majority_label)\n+ ymin = min(y)\n+ if(ymin == 0)\n+ y = y + 1\n- X_under = matrix(0, rows = 0, cols = m)\n- y_under = matrix(0, rows = 0, cols = 1)\n- drop_idx = matrix(0, rows = 0, cols = 1)\n+ # # find the majority labels\n+ label = table(y, 1)\n+ majority_label = as.scalar(rowIndexMax(t(label)))\n- for (i in 1:nrow(X)) {\n- is_link = as.scalar(tomek_links[i, 1])\n- if (is_link == 1) {\n- X_under = rbind(X_under, X[i,])\n- y_under = rbind(y_under, y[i,])\n- drop_idx = rbind(drop_idx, matrix(i, rows = 1, cols = 1))\n- }\n- }\n-}\n+ tomek_links = get_links(X, y, majority_label)\n+ drop_idx = tomek_links * seq(1, nrow(X))\n+ X_under = removeEmpty(target=X, margin=\"rows\", select = (tomek_links == 0))\n+ y_under = removeEmpty(target=y, margin=\"rows\", select = (tomek_links == 0))\n+ drop_idx = removeEmpty(target=drop_idx, margin=\"rows\", select = tomek_links)\n+ if(ymin)\n+ y = y - 1\n-###### END MAIN PART ######\n+}\n-###### UTILS ######\n-# nearest nb function ----------------------------------------------------------\n+# get the nearest neighbour index\nget_nn = function(Matrix[Double] X)\nreturn (Matrix[Double] nn) {\nnn = matrix(0, rows = nrow(X), cols = 1)\n@@ -75,27 +69,13 @@ get_nn = function(Matrix[Double] X)\n}\n}\n-# find tomek link function ----------------------------------------------------\n+# find the tomek links\nget_links = function(Matrix[Double] X, Matrix[Double] y, double majority_label)\nreturn (Matrix[Double] tomek_links) {\n- tomek_links = matrix(0, rows = nrow(X), cols = 1)\nnn = get_nn(X)\n-\n- for (index in 1:nrow(X)) {\n- # this is a tomek link according to R: ubTomek https://rdrr.io/cran/unbalanced/src/R/ubTomek.R\n- # other sources define it as a pair of mutual nearest neighbor\n- # where exactly one endpoint has the majority label\n-\n- nn_index = as.scalar(nn[index, 1])\n- label = as.scalar(y[index, 1])\n- nn_label = as.scalar(y[nn_index, 1])\n-\n- if (label != majority_label) {\n- if (nn_label == majority_label) {\n- tomek_links[nn_index, 1] = 1\n- }\n- }\n- }\n+ perm = table(seq(1, nrow(y)), nn, nrow(y), nrow(y))\n+ nn_labels = perm %*% y\n+ links = (y != majority_label) & (nn_labels == majority_label)\n+ tomek_links = (table(nn, 1, links, nrow(y), 1) > 0)\n}\n-###### END UTILS ######\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/matrix/data/LibMatrixBincell.java", "new_path": "src/main/java/org/apache/sysds/runtime/matrix/data/LibMatrixBincell.java", "diff": "@@ -28,6 +28,8 @@ import java.util.concurrent.ExecutorService;\nimport java.util.concurrent.Future;\nimport org.apache.commons.lang.NotImplementedException;\n+import org.apache.commons.logging.Log;\n+import org.apache.commons.logging.LogFactory;\nimport org.apache.sysds.runtime.DMLRuntimeException;\nimport org.apache.sysds.runtime.data.DenseBlock;\nimport org.apache.sysds.runtime.data.SparseBlock;\n@@ -67,8 +69,9 @@ import org.apache.sysds.runtime.util.UtilFunctions;\n* and sparse-unsafe operations.\n*\n*/\n-public class LibMatrixBincell\n-{\n+public class LibMatrixBincell {\n+\n+ private static final Log LOG = LogFactory.getLog(LibMatrixBincell.class.getName());\nprivate static final long PAR_NUMCELL_THRESHOLD2 = 16*1024; //Min 16K elements\npublic enum BinaryAccessType {\n@@ -244,22 +247,23 @@ public class LibMatrixBincell\n* @param m1ret result matrix updated in place\n* @param m2 matrix block the other matrix to take values from\n* @param op binary operator the operator that is placed in the middle of m1ret and m2\n+ * @return The same pointer to m1ret argument, and the updated result.\n*/\n- public static void bincellOpInPlace(MatrixBlock m1ret, MatrixBlock m2, BinaryOperator op) {\n- bincellOpInPlaceRight(m1ret, m2, op);\n+ public static MatrixBlock bincellOpInPlace(MatrixBlock m1ret, MatrixBlock m2, BinaryOperator op) {\n+ return bincellOpInPlaceRight(m1ret, m2, op);\n}\n/**\n- *\n- * right side operations, updating the m1 matrix with like:\n+ * Right side operations, updating the m1 matrix like:\n*\n* m1ret op m2\n*\n* @param m1ret result matrix updated in place\n* @param m2 matrix block the other matrix to take values from\n* @param op binary operator the operator that is placed in the middle of m1ret and m2\n+ * @return The result MatrixBlock (same object pointer to m1ret argument)\n*/\n- public static void bincellOpInPlaceRight(MatrixBlock m1ret, MatrixBlock m2, BinaryOperator op) {\n+ public static MatrixBlock bincellOpInPlaceRight(MatrixBlock m1ret, MatrixBlock m2, BinaryOperator op) {\n//execute binary cell operations\nif(op.sparseSafe || isSparseSafeDivide(op, m2))\nsafeBinaryInPlace(m1ret, m2, op);\n@@ -270,27 +274,60 @@ public class LibMatrixBincell\n//(no additional memory requirements)\nif( m1ret.isEmptyBlock(false) )\nm1ret.examSparsity();\n+ return m1ret;\n}\n/**\n- *\n- * right side operations, updating the m1 matrix with like:\n+ * Left side operations, updating the m1 matrix like:\n*\n* m2 op m1ret\n*\n* @param m1ret result matrix updated in place\n* @param m2 matrix block the other matrix to take values from\n* @param op binary operator the operator that is placed in the middle of m1ret and m2\n+ * @return The result MatrixBlock (same object pointer to m1ret argument)\n*/\n- public static void bincellOpInPlaceLeft(MatrixBlock m1ret, MatrixBlock m2, BinaryOperator op) {\n- if(m1ret.isInSparseFormat() || m2.isInSparseFormat())\n- throw new NotImplementedException(\"Not implemented sparse inplace left binaryOperator for matrixBlocks\");\n-\n+ public static MatrixBlock bincellOpInPlaceLeft(MatrixBlock m1ret, MatrixBlock m2, BinaryOperator op) {\n+ final int nRows = m1ret.getNumRows();\n+ final int nCols = m1ret.getNumColumns();\n+ if(m1ret.isInSparseFormat()){\n+ // not doing in place, since the m1ret is in sparse format, and m2 might make it dense.\n+ // this is not ideal either, but makes it work\n+ LOG.warn(\"Inefficient bincell op in place left, because output is materialized in new matrix\");\n+ MatrixBlock right = new MatrixBlock(nRows, nCols, true);\n+ right.copyShallow(m1ret);\n+ m1ret.cleanupBlock(true, true);\n+ bincellOp(m2, right, m1ret, op);\n+ return m1ret;\n+ }\n+\n+ // m1ret is dense:\nfinal double[] retV = m1ret.getDenseBlockValues();\n- final double[] m2V = m2.getDenseBlockValues();\n-\n- final int size = m2.getNumColumns() * m2.getNumRows();\nfinal ValueFunction f = op.fn;\n+\n+ if(m2.isInSparseFormat() && op.sparseSafe) {\n+ final SparseBlock sb = m2.getSparseBlock();\n+ for(int row = 0; row < nRows; row++){\n+ if(sb.isEmpty(row)){\n+ continue;\n+ }\n+ final int apos = sb.pos(row);\n+ final int alen = sb.size(row) + apos;\n+ final int[] aix = sb.indexes(row);\n+ final double[] aval = sb.values(row);\n+ final int offsetV = row * nCols;\n+ for(int j = apos; j < alen; j++){\n+ final int idx = offsetV + aix[j];\n+ retV[idx] = f.execute(aval[j], retV[idx]);\n+ }\n+ }\n+ }\n+ else if(m2.isInSparseFormat()){\n+ throw new NotImplementedException(\"Not implemented left bincell in place unsafe operations\");\n+ }\n+ else{\n+ final double[] m2V = m2.getDenseBlockValues();\n+ final int size = nRows * nCols;\nfor(int i = 0; i < size; i++ ){\nretV[i] = f.execute(m2V[i], retV[i]);\n}\n@@ -298,6 +335,8 @@ public class LibMatrixBincell\nif( m1ret.isEmptyBlock(false) )\nm1ret.examSparsity();\n}\n+ return m1ret;\n+ }\npublic static BinaryAccessType getBinaryAccessType(MatrixBlock m1, MatrixBlock m2)\n{\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/matrix/data/LibMatrixAgg.java", "new_path": "src/main/java/org/apache/sysds/runtime/matrix/data/LibMatrixAgg.java", "diff": "@@ -1639,7 +1639,9 @@ public class LibMatrixAgg\nout.quickSetValue(2, j, in.rlen); //count\nbreak;\n}\n-\n+ case CUM_SUM_PROD:{\n+ break;\n+ }\ndefault:\nthrow new DMLRuntimeException(\"Unsupported aggregation type: \"+optype);\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/component/matrix/CumsumprodTest.java", "new_path": "src/test/java/org/apache/sysds/test/component/matrix/CumsumprodTest.java", "diff": "@@ -62,4 +62,13 @@ public class CumsumprodTest {\nMatrixBlock B = A.unaryOperations(uop, new MatrixBlock());\nassertEquals(1000, B.getNumRows());\n}\n+\n+ @Test\n+ public void testCumsumprodEmpty() {\n+ MatrixBlock A = MatrixBlock.randOperations(1000, 2, 0.00, 0, 10, \"uniform\", 7);\n+ A = new MatrixBlock(A, SparseBlock.Type.MCSR, true);\n+ UnaryOperator uop = new UnaryOperator(Builtin.getBuiltinFnObject(\"ucumk+*\"), 1, false);\n+ MatrixBlock B = A.unaryOperations(uop, new MatrixBlock());\n+ assertEquals(1000, B.getNumRows());\n+ }\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/context/ExecutionContext.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/context/ExecutionContext.java", "diff": "@@ -306,6 +306,10 @@ public class ExecutionContext {\nreturn getMatrixObject(varName).acquireRead();\n}\n+ public MatrixBlock getMatrixInput(CPOperand input) {\n+ return getMatrixObject(input.getName()).acquireRead();\n+ }\n+\n/**\n* Pins a matrix variable into memory and returns the internal matrix block.\n*\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/QuaternaryWCeMMFEDInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/QuaternaryWCeMMFEDInstruction.java", "diff": "@@ -67,7 +67,7 @@ public class QuaternaryWCeMMFEDInstruction extends QuaternaryFEDInstruction\nif(qop.hasFourInputs()) {\neps = (_input4.getDataType() == DataType.SCALAR) ?\nec.getScalarInput(_input4) :\n- new DoubleObject(ec.getMatrixInput(_input4.getName()).quickGetValue(0, 0));\n+ new DoubleObject(ec.getMatrixInput(_input4).quickGetValue(0, 0));\n}\nif(X.isFederated()) {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/QuaternaryWDivMMFEDInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/QuaternaryWDivMMFEDInstruction.java", "diff": "@@ -37,11 +37,11 @@ import org.apache.sysds.runtime.DMLRuntimeException;\nimport org.apache.sysds.runtime.instructions.cp.CPOperand;\nimport org.apache.sysds.runtime.instructions.cp.DoubleObject;\nimport org.apache.sysds.runtime.instructions.cp.ScalarObject;\n-import org.apache.sysds.runtime.matrix.operators.Operator;\nimport org.apache.sysds.runtime.matrix.operators.QuaternaryOperator;\nimport java.util.ArrayList;\nimport java.util.concurrent.Future;\n+import java.util.stream.IntStream;\npublic class QuaternaryWDivMMFEDInstruction extends QuaternaryFEDInstruction\n{\n@@ -61,31 +61,34 @@ public class QuaternaryWDivMMFEDInstruction extends QuaternaryFEDInstruction\n* @param opcode ...\n* @param instruction_str ...\n*/\n- protected QuaternaryWDivMMFEDInstruction(Operator operator,\n+\n+ private QuaternaryOperator _qop;\n+\n+ protected QuaternaryWDivMMFEDInstruction(QuaternaryOperator operator,\nCPOperand in1, CPOperand in2, CPOperand in3, CPOperand in4, CPOperand out, String opcode, String instruction_str)\n{\nsuper(FEDType.Quaternary, operator, in1, in2, in3, in4, out, opcode, instruction_str);\n+ _qop = operator;\n}\n@Override\npublic void processInstruction(ExecutionContext ec)\n{\n- QuaternaryOperator qop = (QuaternaryOperator) _optr;\n- final WDivMMType wdivmm_type = qop.wtype3;\n+ final WDivMMType wdivmm_type = _qop.wtype3;\nMatrixObject X = ec.getMatrixObject(input1);\nMatrixObject U = ec.getMatrixObject(input2);\nMatrixObject V = ec.getMatrixObject(input3);\nScalarObject eps = null;\nMatrixObject MX = null;\n- if(qop.hasFourInputs()) {\n+ if(_qop.hasFourInputs()) {\nif(wdivmm_type == WDivMMType.MULT_MINUS_4_LEFT || wdivmm_type == WDivMMType.MULT_MINUS_4_RIGHT) {\nMX = ec.getMatrixObject(_input4);\n}\nelse {\neps = (_input4.getDataType() == DataType.SCALAR) ?\nec.getScalarInput(_input4) :\n- new DoubleObject(ec.getMatrixInput(_input4.getName()).quickGetValue(0, 0));\n+ new DoubleObject(ec.getMatrixInput(_input4).quickGetValue(0, 0));\n}\n}\n@@ -93,7 +96,7 @@ public class QuaternaryWDivMMFEDInstruction extends QuaternaryFEDInstruction\nFederationMap fedMap = X.getFedMapping();\nArrayList<FederatedRequest[]> frSliced = new ArrayList<>();\nArrayList<FederatedRequest> frB = new ArrayList<>(); // FederatedRequests of broadcasts\n- long[] varNewIn = new long[qop.hasFourInputs() ? 4 : 3];\n+ long[] varNewIn = new long[_qop.hasFourInputs() ? 4 : 3];\nvarNewIn[0] = fedMap.getID();\nif(X.isFederated(FType.ROW)) { // row partitioned X\n@@ -151,21 +154,26 @@ public class QuaternaryWDivMMFEDInstruction extends QuaternaryFEDInstruction\n}\nFederatedRequest frComp = FederationUtils.callInstruction(instString, output,\n- qop.hasFourInputs() ? new CPOperand[]{input1, input2, input3, _input4}\n+ _qop.hasFourInputs() ? new CPOperand[]{input1, input2, input3, _input4}\n: new CPOperand[]{input1, input2, input3}, varNewIn);\n// get partial results from federated workers\n- FederatedRequest frGet = new FederatedRequest(RequestType.GET_VAR, frComp.getID());\n+ FederatedRequest frGet = null;\nArrayList<FederatedRequest> frC = new ArrayList<>();\n+ if((wdivmm_type.isLeft() && X.isFederated(FType.ROW))\n+ || (wdivmm_type.isRight() && X.isFederated(FType.COL))) { // output needs local aggregation\n+ frGet = new FederatedRequest(RequestType.GET_VAR, frComp.getID());\nfrC.add(fedMap.cleanup(getTID(), frComp.getID()));\n+ }\nfor(FederatedRequest[] frS : frSliced)\nfrC.add(fedMap.cleanup(getTID(), frS[0].getID()));\nfor(FederatedRequest fr : frB)\nfrC.add(fedMap.cleanup(getTID(), fr.getID()));\n- FederatedRequest[] frAll = ArrayUtils.addAll(ArrayUtils.addAll(\n- frB.toArray(new FederatedRequest[0]), frComp, frGet),\n+ FederatedRequest[] frAll = ArrayUtils.addAll(frGet == null ?\n+ ArrayUtils.addAll(frB.toArray(new FederatedRequest[0]), frComp) :\n+ ArrayUtils.addAll(frB.toArray(new FederatedRequest[0]), frComp, frGet),\nfrC.toArray(new FederatedRequest[0]));\n// execute federated instructions\n@@ -174,14 +182,13 @@ public class QuaternaryWDivMMFEDInstruction extends QuaternaryFEDInstruction\ngetTID(), true, frSliced.toArray(new FederatedRequest[0][]), frAll);\nif((wdivmm_type.isLeft() && X.isFederated(FType.ROW))\n- || (wdivmm_type.isRight() && X.isFederated(FType.COL))) {\n+ || (wdivmm_type.isRight() && X.isFederated(FType.COL))) { // local aggregation\n// aggregate partial results from federated responses\nAggregateUnaryOperator aop = InstructionUtils.parseBasicAggregateUnaryOperator(\"uak+\");\nec.setMatrixOutput(output.getName(), FederationUtils.aggMatrix(aop, response, fedMap));\n}\nelse if(wdivmm_type.isLeft() || wdivmm_type.isRight() || wdivmm_type.isBasic()) {\n- // bind partial results from federated responses\n- ec.setMatrixOutput(output.getName(), FederationUtils.bind(response, false));\n+ setFederatedOutput(X, U, V, ec, frComp.getID());\n}\nelse {\nthrow new DMLRuntimeException(\"Federated WDivMM only supported for BASIC, LEFT or RIGHT variants.\");\n@@ -192,5 +199,53 @@ public class QuaternaryWDivMMFEDInstruction extends QuaternaryFEDInstruction\n+ X.isFederated() + \", \" + U.isFederated() + \", \" + V.isFederated() + \")\");\n}\n}\n+\n+ /**\n+ * Set the federated output according to the output data charactersitics of\n+ * the different wdivmm types\n+ */\n+ private void setFederatedOutput(MatrixObject X, MatrixObject U, MatrixObject V, ExecutionContext ec, long fedMapID) {\n+ final WDivMMType wdivmm_type = _qop.wtype3;\n+ MatrixObject out = ec.getMatrixObject(output);\n+ FederationMap outFedMap = X.getFedMapping().copyWithNewID(fedMapID);\n+\n+ long rows = -1;\n+ long cols = -1;\n+ if(wdivmm_type.isBasic()) {\n+ // BASIC: preserve dimensions of X\n+ rows = X.getNumRows();\n+ cols = X.getNumColumns();\n+ }\n+ else if(wdivmm_type.isLeft()) {\n+ // LEFT: nrows of transposed X, ncols of U\n+ rows = X.getNumColumns();\n+ cols = U.getNumColumns();\n+ outFedMap = modifyFedRanges(outFedMap.transpose(), cols, 1);\n+ }\n+ else if(wdivmm_type.isRight()) {\n+ // RIGHT: nrows of X, ncols of V\n+ rows = X.getNumRows();\n+ cols = V.getNumColumns();\n+ outFedMap = modifyFedRanges(outFedMap, cols, 1);\n+ }\n+ out.setFedMapping(outFedMap);\n+ out.getDataCharacteristics().set(rows, cols, (int) X.getBlocksize());\n}\n+ /**\n+ * Takes the federated mapping and sets one dimension of all federated ranges\n+ * to the specified value.\n+ *\n+ * @param fedMap the original federated mapping\n+ * @param value long value for setting the dimension\n+ * @param dim indicates if the row (0) or column (1) dimension should be set to value\n+ * @return FederationMap with the modified federated ranges\n+ */\n+ private static FederationMap modifyFedRanges(FederationMap fedMap, long value, int dim) {\n+ IntStream.range(0, fedMap.getFederatedRanges().length).forEach(i -> {\n+ fedMap.getFederatedRanges()[i].setBeginDim(dim, 0);\n+ fedMap.getFederatedRanges()[i].setEndDim(dim, value);\n+ });\n+ return fedMap;\n+ }\n+}\n" } ]

[ { "change_type": "MODIFY", "old_path": "docs/site/builtins-reference.md", "new_path": "docs/site/builtins-reference.md", "diff": "@@ -79,6 +79,7 @@ limitations under the License.\n* [`steplm`-Function](#steplm-function)\n* [`tomekLink`-Function](#tomekLink-function)\n* [`toOneHot`-Function](#toOneHOt-function)\n+ * [`tSNE`-Function](#tSNE-function)\n* [`winsorize`-Function](#winsorize-function)\n* [`xgboost`-Function](#xgboost-function)\n@@ -2176,6 +2177,40 @@ X = round(rand(rows = 10, cols = 10, min = 1, max = numClasses))\ny = toOneHot(X,numClasses)\n```\n+## `tSNE`-Function\n+\n+The `tSNE`-function performs dimensionality reduction using tSNE algorithm based on the paper: Visualizing Data using t-SNE, Maaten et. al.\n+\n+### Usage\n+\n+```r\n+tSNE(X, reduced_dims, perplexity, lr, momentum, max_iter, seed, is_verbose)\n+```\n+\n+### Arguments\n+\n+| Name | Type | Default | Description |\n+| :----------- | :------------- | -------- | :---------- |\n+| X | Matrix[Double] | required | Data Matrix of shape (number of data points, input dimensionality) |\n+| reduced_dims | Integer | 2 | Output dimensionality |\n+| perplexity | Integer | 30 | Perplexity Parameter |\n+| lr | Double | 300. | Learning rate |\n+| momentum | Double | 0.9 | Momentum Parameter |\n+| max_iter | Integer | 1000 | Number of iterations |\n+| seed | Integer | -1 | The seed used for initial values. If set to -1 random seeds are selected. |\n+| is_verbose | Boolean | FALSE | Print debug information |\n+### Returns\n+\n+| Type | Description |\n+| :------------- | :---------- |\n+| Matrix[Double] | Data Matrix of shape (number of data points, reduced_dims) |\n+\n+### Example\n+\n+```r\n+X = rand(rows = 100, cols = 10, min = -10, max = 10))\n+Y = tSNE(X)\n+```\n## `winsorize`-Function\n" }, { "change_type": "ADD", "old_path": null, "new_path": "scripts/builtin/tSNE.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+# This function performs dimensionality reduction using tSNE algorithm based on\n+# the paper: Visualizing Data using t-SNE, Maaten et. al.\n+\n+# INPUT PARAMETERS:\n+# ----------------------------------------------------------------------------\n+# NAME TYPE DEFAULT MEANING\n+# ----------------------------------------------------------------------------\n+# X Double --- Data Matrix of shape\n+# (number of data points, input dimensionality)\n+# reduced_dims Integer 2 Output dimensionality\n+# perplexity Integer 30 Perplexity Parameter\n+# lr Double 300. Learning rate\n+# momentum Double 0.9 Momentum Parameter\n+# max_iter Integer 1000 Number of iterations\n+# seed Integer -1 The seed used for initial values.\n+# If set to -1 random seeds are selected.\n+# is_verbose Boolean FALSE Print debug information\n+#\n+#\n+# RETURN VALUES\n+# ----------------------------------------------------------------------------\n+# NAME TYPE DEFAULT MEANING\n+# ----------------------------------------------------------------------------\n+# Y Matrix --- Data Matrix of shape (number of data points, reduced_dims)\n+# ----------------------------------------------------------------------------\n+\n+\n+m_tSNE = function(Matrix[Double] X, Integer reduced_dims = 2, Integer perplexity = 30,\n+ Double lr = 300., Double momentum = 0.9, Integer max_iter = 1000, Integer seed = -1, Boolean is_verbose = FALSE)\n+ return(Matrix[Double] Y)\n+{\n+ d = reduced_dims\n+ n = nrow(X)\n+\n+ P = x2p(X, perplexity, is_verbose)\n+ P = P*4\n+ Y = rand(rows=n, cols=d, pdf=\"normal\", seed=seed)\n+ dY = matrix(0, rows=n, cols=d)\n+ C = matrix(0, rows=max_iter/100, cols=1)\n+ ZERODIAG = (diag(matrix(-1, rows=n, cols=1)) + 1)\n+\n+ D = matrix(0, rows=n, cols=n)\n+ Z = matrix(0, rows=n, cols=n)\n+ Q = matrix(0, rows=n, cols=n)\n+ W = matrix(0, rows=n, cols=n)\n+\n+ if(is_verbose)\n+ print(\"starting loop....\")\n+\n+ for (itr in 1:max_iter) {\n+ D = distance_matrix(Y)\n+ Z = 1/(D + 1)\n+ Z = Z * ZERODIAG\n+ Q = Z/sum(Z)\n+ W = (P - Q)*Z\n+ sumW = rowSums(W)\n+ g = Y * sumW - W %*% Y\n+ dY = momentum*dY - lr*g\n+ Y = Y + dY\n+ Y = Y - colMeans(Y)\n+\n+ if (itr%%100 == 0) {\n+ C[itr/100,] = sum(P * log(pmax(P, 1e-12) / pmax(Q, 1e-12)))\n+ }\n+ if (itr == 100) {\n+ P = P/4\n+ }\n+ }\n+}\n+\n+distance_matrix = function(matrix[double] X)\n+ return (matrix[double] out)\n+{\n+ # TODO consolidate with dist() builtin, but with\n+ # better way of obtaining the diag from\n+ n = nrow(X)\n+ s = rowSums(X * X)\n+ out = - 2*X %*% t(X) + s + t(s)\n+}\n+\n+\n+x2p = function(matrix[double] X, double perplexity, Boolean is_verbose = FALSE)\n+return(matrix[double] P)\n+{\n+ if(is_verbose)\n+ print(\"x2p....\")\n+ tol = 1.0e-5\n+ INF = 1.0e20\n+ n = nrow(X)\n+ if(is_verbose)\n+ print(n)\n+ D = distance_matrix(X)\n+\n+ P = matrix(0, rows=n, cols=n)\n+ beta = matrix(1, rows=n, cols=1)\n+ betamax = matrix(INF, rows=n, cols=1)\n+ betamin = matrix(0, rows=n, cols=1)\n+ Hdiff = matrix(INF, rows=n, cols=1)\n+ logU = log(perplexity)\n+\n+ ZERODIAG = (diag(matrix(-1, rows=n, cols=1)) + 1)\n+ itr = 1\n+ while (mean(abs(Hdiff)) > tol & itr < 50) {\n+ P = exp(-D * beta)\n+ P = P * ZERODIAG\n+ sum_Pi = rowSums(P)\n+ W = rowSums(P * D)\n+ Ws = W/sum_Pi\n+ H = log(sum_Pi) + beta * Ws\n+ P = P/sum_Pi\n+ Hdiff = H - logU\n+\n+ Hpos = (Hdiff >= 0)\n+ Hneg = (Hdiff < 0)\n+ betamin = Hneg*betamin + Hpos*beta\n+ betamax = Hpos*betamax + Hneg*beta\n+ beta = 2*Hpos*(betamax == INF)*beta +\n+ Hpos*(betamax != INF)*(beta + betamax)/2 +\n+ Hneg*(beta + betamin)/2\n+\n+ itr = itr + 1\n+ }\n+\n+ P = P + t(P)\n+ P = P / sum(P)\n+ if(is_verbose)\n+ print(\"x2p finishing....\")\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/common/Builtins.java", "new_path": "src/main/java/org/apache/sysds/common/Builtins.java", "diff": "@@ -260,6 +260,7 @@ public enum Builtins {\nTOMEKLINK(\"tomeklink\", true),\nTRACE(\"trace\", false),\nTRANS(\"t\", false),\n+ TSNE(\"tSNE\", true),\nTYPEOF(\"typeof\", false),\nUNIVAR(\"univar\", true),\nVAR(\"var\", false),\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/util/AutoDiff.java", "new_path": "src/main/java/org/apache/sysds/runtime/util/AutoDiff.java", "diff": "@@ -48,7 +48,7 @@ public class AutoDiff {\npublic static ListObject getBackward(MatrixObject mo, ArrayList<Data> lineage, ExecutionContext adec) {\n- ArrayList<String> names = new ArrayList<String>();\n+ ArrayList<String> names = new ArrayList<>();\n// parse the lineage and take the number of instructions as for each instruction there is separate hop DAG\nString lin = lineage.get(0).toString();\n// get rid of foo flag\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysds/test/functions/builtin/BuiltinTSNETest.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysds.test.functions.builtin;\n+\n+import org.apache.sysds.common.Types.ExecMode;\n+import org.apache.sysds.common.Types.ExecType;\n+import org.apache.sysds.runtime.matrix.data.MatrixValue;\n+import org.apache.sysds.runtime.matrix.data.MatrixValue.CellIndex;\n+import org.apache.sysds.test.AutomatedTestBase;\n+import org.apache.sysds.test.TestConfiguration;\n+import org.junit.Assert;\n+import org.junit.Test;\n+\n+import java.io.IOException;\n+import java.util.HashMap;\n+import java.util.Map.Entry;\n+\n+public class BuiltinTSNETest extends AutomatedTestBase\n+{\n+ private final static String TEST_NAME = \"tSNE\";\n+ private final static String TEST_DIR = \"functions/builtin/\";\n+ private static final String TEST_CLASS_DIR = TEST_DIR + BuiltinTSNETest.class.getSimpleName() + \"/\";\n+\n+ @Override\n+ public void setUp() {\n+ addTestConfiguration(TEST_NAME,new TestConfiguration(TEST_CLASS_DIR, TEST_NAME,new String[]{\"B\"}));\n+ }\n+\n+ @Test\n+ public void testTSNECP() throws IOException {\n+ runTSNETest(2, 30, 300.,\n+ 0.9, 1000, 42, \"FALSE\", ExecType.CP);\n+ }\n+\n+ private void runTSNETest(Integer reduced_dims, Integer perplexity, Double lr,\n+ Double momentum, Integer max_iter, Integer seed, String is_verbose, ExecType instType)\n+ throws IOException\n+ {\n+ ExecMode platformOld = setExecMode(instType);\n+\n+ try\n+ {\n+ loadTestConfiguration(getTestConfiguration(TEST_NAME));\n+\n+ String HOME = SCRIPT_DIR + TEST_DIR;\n+ fullDMLScriptName = HOME + TEST_NAME + \".dml\";\n+ programArgs = new String[]{\n+ \"-nvargs\", \"X=\" + input(\"X\"), \"Y=\" + output(\"Y\"),\n+ \"reduced_dims=\" + reduced_dims,\n+ \"perplexity=\" + perplexity,\n+ \"lr=\" + lr,\n+ \"momentum=\" + momentum,\n+ \"max_iter=\" + max_iter,\n+ \"seed=\" + seed,\n+ \"is_verbose=\" + is_verbose};\n+\n+ // The Input values are calculated using the following R script:\n+ // TODO create via dml operations, avoid inlining data\n+ // library(Rtsne)\n+ // set.seed(42)\n+ // iris_unique <- unique(iris)\n+ // iris_matrix <- as.matrix(iris_unique[,1:4])\n+ // X <- normalize_input(iris_matrix) # the values used for the test\n+\n+ // Input\n+ double[][] X = {{-0.23599574, 0.13972311, -0.74547391, -0.31565495},\n+ {-0.29946752, -0.01895634, -0.74547391, -0.31565495},\n+ {-0.36293930, 0.04451544, -0.77720980, -0.31565495},\n+ {-0.39467519, 0.01277955, -0.71373802, -0.31565495},\n+ {-0.26773163, 0.17145900, -0.74547391, -0.31565495},\n+ {-0.14078807, 0.26666667, -0.65026624, -0.25218317},\n+ {-0.39467519, 0.10798722, -0.74547391, -0.28391906},\n+ {-0.26773163, 0.10798722, -0.71373802, -0.31565495},\n+ {-0.45814696, -0.05069223, -0.74547391, -0.31565495},\n+ {-0.29946752, 0.01277955, -0.71373802, -0.34739084},\n+ {-0.14078807, 0.20319489, -0.71373802, -0.31565495},\n+ {-0.33120341, 0.10798722, -0.68200213, -0.31565495},\n+ {-0.33120341, -0.01895634, -0.74547391, -0.34739084},\n+ {-0.48988285, -0.01895634, -0.84068158, -0.34739084},\n+ {-0.01384452, 0.29840256, -0.80894569, -0.31565495},\n+ {-0.04558040, 0.42534611, -0.71373802, -0.25218317},\n+ {-0.14078807, 0.26666667, -0.77720980, -0.25218317},\n+ {-0.23599574, 0.13972311, -0.74547391, -0.28391906},\n+ {-0.04558040, 0.23493078, -0.65026624, -0.28391906},\n+ {-0.23599574, 0.23493078, -0.71373802, -0.28391906},\n+ {-0.14078807, 0.10798722, -0.65026624, -0.31565495},\n+ {-0.23599574, 0.20319489, -0.71373802, -0.25218317},\n+ {-0.39467519, 0.17145900, -0.87241747, -0.31565495},\n+ {-0.23599574, 0.07625133, -0.65026624, -0.22044728},\n+ {-0.33120341, 0.10798722, -0.58679446, -0.31565495},\n+ {-0.26773163, -0.01895634, -0.68200213, -0.31565495},\n+ {-0.26773163, 0.10798722, -0.68200213, -0.25218317},\n+ {-0.20425985, 0.13972311, -0.71373802, -0.31565495},\n+ {-0.20425985, 0.10798722, -0.74547391, -0.31565495},\n+ {-0.36293930, 0.04451544, -0.68200213, -0.31565495},\n+ {-0.33120341, 0.01277955, -0.68200213, -0.31565495},\n+ {-0.14078807, 0.10798722, -0.71373802, -0.25218317},\n+ {-0.20425985, 0.33013845, -0.71373802, -0.34739084},\n+ {-0.10905218, 0.36187433, -0.74547391, -0.31565495},\n+ {-0.29946752, 0.01277955, -0.71373802, -0.31565495},\n+ {-0.26773163, 0.04451544, -0.80894569, -0.31565495},\n+ {-0.10905218, 0.13972311, -0.77720980, -0.31565495},\n+ {-0.29946752, 0.17145900, -0.74547391, -0.34739084},\n+ {-0.45814696, -0.01895634, -0.77720980, -0.31565495},\n+ {-0.23599574, 0.10798722, -0.71373802, -0.31565495},\n+ {-0.26773163, 0.13972311, -0.77720980, -0.28391906},\n+ {-0.42641108, -0.24110756, -0.77720980, -0.28391906},\n+ {-0.45814696, 0.04451544, -0.77720980, -0.31565495},\n+ {-0.26773163, 0.13972311, -0.68200213, -0.18871140},\n+ {-0.23599574, 0.23493078, -0.58679446, -0.25218317},\n+ {-0.33120341, -0.01895634, -0.74547391, -0.28391906},\n+ {-0.23599574, 0.23493078, -0.68200213, -0.31565495},\n+ {-0.39467519, 0.04451544, -0.74547391, -0.31565495},\n+ {-0.17252396, 0.20319489, -0.71373802, -0.31565495},\n+ {-0.26773163, 0.07625133, -0.74547391, -0.31565495},\n+ {0.36698616, 0.04451544, 0.30181044, 0.06517572},\n+ {0.17657082, 0.04451544, 0.23833866, 0.09691161},\n+ {0.33525027, 0.01277955, 0.36528222, 0.09691161},\n+ {-0.10905218, -0.24110756, 0.07965921, 0.03343983},\n+ {0.20830671, -0.08242812, 0.27007455, 0.09691161},\n+ {-0.04558040, -0.08242812, 0.23833866, 0.03343983},\n+ {0.14483493, 0.07625133, 0.30181044, 0.12864750},\n+ {-0.29946752, -0.20937167, -0.14249201, -0.06176784},\n+ {0.24004260, -0.05069223, 0.27007455, 0.03343983},\n+ {-0.20425985, -0.11416400, 0.04792332, 0.06517572},\n+ {-0.26773163, -0.33631523, -0.07902023, -0.06176784},\n+ {0.01789137, -0.01895634, 0.14313099, 0.09691161},\n+ {0.04962726, -0.27284345, 0.07965921, -0.06176784},\n+ {0.08136315, -0.05069223, 0.30181044, 0.06517572},\n+ {-0.07731629, -0.05069223, -0.04728435, 0.03343983},\n+ {0.27177849, 0.01277955, 0.20660277, 0.06517572},\n+ {-0.07731629, -0.01895634, 0.23833866, 0.09691161},\n+ {-0.01384452, -0.11416400, 0.11139510, -0.06176784},\n+ {0.11309904, -0.27284345, 0.23833866, 0.09691161},\n+ {-0.07731629, -0.17763578, 0.04792332, -0.03003195},\n+ {0.01789137, 0.04451544, 0.33354633, 0.19211928},\n+ {0.08136315, -0.08242812, 0.07965921, 0.03343983},\n+ {0.14483493, -0.17763578, 0.36528222, 0.09691161},\n+ {0.08136315, -0.08242812, 0.30181044, 0.00170394},\n+ {0.17657082, -0.05069223, 0.17486688, 0.03343983},\n+ {0.24004260, -0.01895634, 0.20660277, 0.06517572},\n+ {0.30351438, -0.08242812, 0.33354633, 0.06517572},\n+ {0.27177849, -0.01895634, 0.39701810, 0.16038339},\n+ {0.04962726, -0.05069223, 0.23833866, 0.09691161},\n+ {-0.04558040, -0.14589989, -0.07902023, -0.06176784},\n+ {-0.10905218, -0.20937167, 0.01618743, -0.03003195},\n+ {-0.10905218, -0.20937167, -0.01554846, -0.06176784},\n+ {-0.01384452, -0.11416400, 0.04792332, 0.00170394},\n+ {0.04962726, -0.11416400, 0.42875399, 0.12864750},\n+ {-0.14078807, -0.01895634, 0.23833866, 0.09691161},\n+ {0.04962726, 0.10798722, 0.23833866, 0.12864750},\n+ {0.27177849, 0.01277955, 0.30181044, 0.09691161},\n+ {0.14483493, -0.24110756, 0.20660277, 0.03343983},\n+ {-0.07731629, -0.01895634, 0.11139510, 0.03343983},\n+ {-0.10905218, -0.17763578, 0.07965921, 0.03343983},\n+ {-0.10905218, -0.14589989, 0.20660277, 0.00170394},\n+ {0.08136315, -0.01895634, 0.27007455, 0.06517572},\n+ {-0.01384452, -0.14589989, 0.07965921, 0.00170394},\n+ {-0.26773163, -0.24110756, -0.14249201, -0.06176784},\n+ {-0.07731629, -0.11416400, 0.14313099, 0.03343983},\n+ {-0.04558040, -0.01895634, 0.14313099, 0.00170394},\n+ {-0.04558040, -0.05069223, 0.14313099, 0.03343983},\n+ {0.11309904, -0.05069223, 0.17486688, 0.03343983},\n+ {-0.23599574, -0.17763578, -0.23769968, -0.03003195},\n+ {-0.04558040, -0.08242812, 0.11139510, 0.03343983},\n+ {0.14483493, 0.07625133, 0.71437700, 0.41427050},\n+ {-0.01384452, -0.11416400, 0.42875399, 0.22385517},\n+ {0.39872204, -0.01895634, 0.68264111, 0.28732694},\n+ {0.14483493, -0.05069223, 0.58743344, 0.19211928},\n+ {0.20830671, -0.01895634, 0.65090522, 0.31906283},\n+ {0.55740149, -0.01895634, 0.90479233, 0.28732694},\n+ {-0.29946752, -0.17763578, 0.23833866, 0.16038339},\n+ {0.46219382, -0.05069223, 0.80958466, 0.19211928},\n+ {0.27177849, -0.17763578, 0.65090522, 0.19211928},\n+ {0.43045793, 0.17145900, 0.74611289, 0.41427050},\n+ {0.20830671, 0.04451544, 0.42875399, 0.25559105},\n+ {0.17657082, -0.11416400, 0.49222577, 0.22385517},\n+ {0.30351438, -0.01895634, 0.55569755, 0.28732694},\n+ {-0.04558040, -0.17763578, 0.39701810, 0.25559105},\n+ {-0.01384452, -0.08242812, 0.42875399, 0.38253461},\n+ {0.17657082, 0.04451544, 0.49222577, 0.35079872},\n+ {0.20830671, -0.01895634, 0.55569755, 0.19211928},\n+ {0.58913738, 0.23493078, 0.93652822, 0.31906283},\n+ {0.58913738, -0.14589989, 1.00000000, 0.35079872},\n+ {0.04962726, -0.27284345, 0.39701810, 0.09691161},\n+ {0.33525027, 0.04451544, 0.61916933, 0.35079872},\n+ {-0.07731629, -0.08242812, 0.36528222, 0.25559105},\n+ {0.58913738, -0.08242812, 0.93652822, 0.25559105},\n+ {0.14483493, -0.11416400, 0.36528222, 0.19211928},\n+ {0.27177849, 0.07625133, 0.61916933, 0.28732694},\n+ {0.43045793, 0.04451544, 0.71437700, 0.19211928},\n+ {0.11309904, -0.08242812, 0.33354633, 0.19211928},\n+ {0.08136315, -0.01895634, 0.36528222, 0.19211928},\n+ {0.17657082, -0.08242812, 0.58743344, 0.28732694},\n+ {0.43045793, -0.01895634, 0.65090522, 0.12864750},\n+ {0.49392971, -0.08242812, 0.74611289, 0.22385517},\n+ {0.65260916, 0.23493078, 0.84132055, 0.25559105},\n+ {0.17657082, -0.08242812, 0.58743344, 0.31906283},\n+ {0.14483493, -0.08242812, 0.42875399, 0.09691161},\n+ {0.08136315, -0.14589989, 0.58743344, 0.06517572},\n+ {0.58913738, -0.01895634, 0.74611289, 0.35079872},\n+ {0.14483493, 0.10798722, 0.58743344, 0.38253461},\n+ {0.17657082, 0.01277955, 0.55569755, 0.19211928},\n+ {0.04962726, -0.01895634, 0.33354633, 0.19211928},\n+ {0.33525027, 0.01277955, 0.52396166, 0.28732694},\n+ {0.27177849, 0.01277955, 0.58743344, 0.38253461},\n+ {0.33525027, 0.01277955, 0.42875399, 0.35079872},\n+ {0.30351438, 0.04451544, 0.68264111, 0.35079872},\n+ {0.27177849, 0.07625133, 0.61916933, 0.41427050},\n+ {0.27177849, -0.01895634, 0.46048988, 0.35079872},\n+ {0.14483493, -0.17763578, 0.39701810, 0.22385517},\n+ {0.20830671, -0.01895634, 0.46048988, 0.25559105},\n+ {0.11309904, 0.10798722, 0.52396166, 0.35079872},\n+ {0.01789137, -0.01895634, 0.42875399, 0.19211928}};\n+\n+ // The reference output was created by using the builtin function with seed 42 and visually inspecting the\n+ // result with the following addition to the above R script:\n+ /*\n+ plot(Y, col = iris_unique$Species)\n+ */\n+\n+ // reference Output\n+ double[][] YReference = {{18.220536548250042, -12.846498524536738},\n+ {15.927903386925026, -14.212023388236792},\n+ {16.769777454402725, -14.867104469807458},\n+ {16.290613410318578, -14.971912325413014},\n+ {18.534108527624923, -13.081965971299278},\n+ {19.46702930119709, -11.107384827606543},\n+ {17.196995022994926, -14.952457676596161},\n+ {17.531360762128234, -13.133905834551287},\n+ {15.996750713161672, -15.670577143806288},\n+ {16.36534147032176, -13.94640444049381},\n+ {19.094349767837077, -11.557657039778153},\n+ {16.909635859249846, -13.432332957158627},\n+ {15.964241411757008, -14.583849627922195},\n+ {16.313709524761837, -16.090269929669734},\n+ {20.285962611966927, -10.881660944862407},\n+ {20.554758173661426, -11.06392329099603},\n+ {19.81906056722687, -11.547188487333667},\n+ {18.156895316378105, -12.705433004326382},\n+ {19.567551886989456, -10.747111880219315},\n+ {19.250142947939032, -12.360380239016678},\n+ {17.98651875041291, -11.373400820175243},\n+ {18.85701724837406, -12.344092918634603},\n+ {15.003071782449434, -15.134107990259263},\n+ {16.906273660742716, -12.081578738297303},\n+ {16.191062837166225, -12.743134302084322},\n+ {15.882718808144244, -13.620171658276757},\n+ {17.162351058283594, -12.52282386909434},\n+ {18.222170807106863, -12.25960056343841},\n+ {17.965249267850382, -12.497678277213915},\n+ {16.61308662972421, -14.338853140723119},\n+ {16.181343729377808, -14.009495314657007},\n+ {18.02370742093665, -11.5469978987216},\n+ {20.11583352615154, -11.971703104743623},\n+ {20.34766483388781, -11.368324149466424},\n+ {16.425063301904068, -13.920590257944497},\n+ {17.456451755948965, -14.167821170017678},\n+ {18.596836114631266, -11.38629820258609},\n+ {18.433917702277277, -13.54006232724752},\n+ {16.220662327912546, -15.705004632184767},\n+ {17.693153385837373, -12.795456099765842},\n+ {18.171476611546456, -13.323284884616521},\n+ {15.440053805476438, -16.14242285042382},\n+ {16.61041486346442, -15.64274785656758},\n+ {17.151462077196808, -11.907401620659614},\n+ {18.626998818277606, -10.74416620713765},\n+ {15.98116134477208, -14.570088706192026},\n+ {19.310896130733955, -12.328582468776172},\n+ {16.638185072116197, -15.041546247833768},\n+ {19.00600656884603, -11.840914144209874},\n+ {17.403065129854806, -13.533995008873486},\n+ {-7.038268225948856, 7.536962272871995},\n+ {-7.3914411713029935, 5.908404973751881},\n+ {-7.496137452637684, 7.80002229833549},\n+ {-6.460592719894593, 0.9062234315612744},\n+ {-7.616915840513189, 6.367666534180176},\n+ {-6.73555772449111, 3.123913125534304},\n+ {-8.080845938411253, 6.030180083529746},\n+ {-6.003581004054341, -0.5426273138978279},\n+ {-7.169789369755973, 6.530569753225136},\n+ {-5.74311471355638, 1.109042979038742},\n+ {-6.159692868338611, -0.47672936959379775},\n+ {-7.224663019809244, 3.2705583465396795},\n+ {-7.80311490665107, 0.8971687420050435},\n+ {-8.171493899007842, 4.945491553039871},\n+ {-6.055076176778027, 0.8068408811777839},\n+ {-6.841744219152401, 6.563990176275155},\n+ {-6.4688066027110995, 3.3600337950542185},\n+ {-7.290016820260793, 1.6539424879073648},\n+ {-9.427559060500517, 3.747497254956788},\n+ {-6.870144371536429, 0.8988782592514124},\n+ {-9.136986152714911, 5.784098425135586},\n+ {-7.68317760808878, 2.43574703565265},\n+ {-9.740448694574955, 5.387219781790028},\n+ {-8.073455284093917, 4.572592063175601},\n+ {-6.941003784668771, 5.2940569383973815},\n+ {-6.92987101320006, 6.2327815077976485},\n+ {-7.48373098301499, 7.341059972461793},\n+ {-8.40332894242311, 8.176817619574457},\n+ {-7.681170155620538, 4.334864702224226},\n+ {-6.956305041180643, 0.2075870963686768},\n+ {-6.655261285197925, 0.5228359022734664},\n+ {-6.666350115623071, 0.2717264575985321},\n+ {-7.0998853298193145, 1.4238408353689687},\n+ {-10.2308151080835, 6.427565359129248},\n+ {-6.01229671664338, 3.082044215406439},\n+ {-8.034938379871582, 5.334134289261261},\n+ {-7.359172540935799, 7.109704156745161},\n+ {-9.171082430966738, 3.5626753890742657},\n+ {-6.535162057220474, 2.312414150179807},\n+ {-6.3789219417924405, 1.2216485793591474},\n+ {-6.067256940518085, 2.1167290391035074},\n+ {-7.812215578205332, 4.875684399693769},\n+ {-7.1374342502995605, 1.4628113062889514},\n+ {-6.070547359306737, -0.5122472965230369},\n+ {-6.46343725557135, 1.979317360428436},\n+ {-6.847076431902512, 2.5392196334816846},\n+ {-6.808114424651112, 2.498147826359703},\n+ {-7.202776337109158, 4.407291480436176},\n+ {-6.008168593188563, -0.6619390315875436},\n+ {-6.831119010530879, 2.045771384669187},\n+ {-12.091830779150229, 11.265359769601078},\n+ {-10.854204065227533, 6.234604499378284},\n+ {-10.722649555263748, 12.399149192535997},\n+ {-11.024368675274824, 9.162144335131195},\n+ {-11.303048023519185, 10.732650608737089},\n+ {-11.09821732195217, 13.779405512286052},\n+ {-5.0247832693311425, 2.1742184328563603},\n+ {-10.619343405836839, 13.298677973987925},\n+ {-11.79686116542718, 9.494394976289728},\n+ {-11.586950451837295, 12.746499445013628},\n+ {-9.748309996567489, 9.564317509433048},\n+ {-10.481918971993704, 8.61651017140811},\n+ {-10.462599559239374, 10.911028525489696},\n+ {-11.130930155766606, 5.964285024103242},\n+ {-11.553937238933786, 6.4713896554450985},\n+ {-10.681883406818072, 10.318460208361941},\n+ {-10.683549192372697, 9.495275009653032},\n+ {-10.78603000302186, 14.25264787910952},\n+ {-11.488550071992856, 13.974915029487743},\n+ {-10.319674004978886, 4.889390024695409},\n+ {-10.860193039087495, 11.683572442191727},\n+ {-11.051151523346409, 5.890167964580233},\n+ {-11.229564353462594, 13.877567340046305},\n+ {-9.673077426769021, 6.894318103609319},\n+ {-10.96913834998139, 11.322206899793283},\n+ {-10.490035372117925, 12.825714370138753},\n+ {-9.462007223927465, 6.4675755622386655},\n+ {-9.51159790297749, 6.337629581532479},\n+ {-11.162432115640609, 9.827386758865119},\n+ {-10.151930853808723, 12.721404715075673},\n+ {-10.604601715753107, 13.20203200422886},\n+ {-10.661315153978798, 14.213091123997359},\n+ {-11.20216179567001, 10.027518789368527},\n+ {-9.545267493844456, 7.1692559024456735},\n+ {-11.159867362151168, 7.8902037332855635},\n+ {-11.088628333620173, 13.451554645466246},\n+ {-11.888514670355377, 10.819305597527286},\n+ {-10.69716594839169, 9.44576929331764},\n+ {-9.388828314326277, 6.02029293546396},\n+ {-10.20160261480168, 11.075164265002408},\n+ {-10.984952128353312, 11.125251232506113},\n+ {-9.734560763746329, 10.70872555677429},\n+ {-11.239599613172867, 11.820790907417845},\n+ {-11.469876956620716, 11.460531594176354},\n+ {-10.037039090761771, 10.470461014159099},\n+ {-10.182738080227073, 7.165787375813321},\n+ {-10.04300898871791, 9.494772412049684},\n+ {-11.912720679251729, 10.441038042708657},\n+ {-10.30731700479772, 6.343742643599125}};\n+\n+ writeInputMatrixWithMTD(\"X\", X, true);\n+\n+ runTest(true, false, null, -1);\n+ HashMap<MatrixValue.CellIndex, Double> dmlFileY = readDMLMatrixFromOutputDir(\"Y\");\n+\n+ // Verifying\n+ for (Entry<CellIndex, Double> entry : dmlFileY.entrySet()) {\n+ MatrixValue.CellIndex key = entry.getKey();\n+ Double value = entry.getValue();\n+ Assert.assertEquals(\"The DML data for cell (\" + key.row + \",\" + key.column + \") '\" + value + \"' is \" +\n+ \"not equal to the expected value '\" + YReference[key.row-1][key.column-1] + \"'\",\n+ YReference[key.row-1][key.column-1], value, 3); //TODO algorithm-level differences?\n+ }\n+ }\n+ finally {\n+ rtplatform = platformOld;\n+ }\n+ }\n+}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/builtin/tSNE.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+X = read($X);\n+Y = tSNE(X, $reduced_dims, $perplexity, $lr, $momentum, $max_iter, $seed, $is_verbose)\n+write(Y, $Y)\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/builtin/dist.dml", "new_path": "scripts/builtin/dist.dml", "diff": "m_dist = function(Matrix[Double] X) return (Matrix[Double] Y) {\nG = X %*% t(X);\n- I = matrix(1, rows = nrow(G), cols = ncol(G));\n- Y = -2 * (G) + (diag(G) * I) + (I * t(diag(G)));\n- Y = sqrt(Y);\n+ Y = sqrt(-2 * G + outer(diag(G), t(diag(G)), \"+\"));\nY = replace(target = Y, pattern=0/0, replacement = 0);\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/api/DMLScript.java", "new_path": "src/main/java/org/apache/sysds/api/DMLScript.java", "diff": "@@ -67,6 +67,7 @@ import org.apache.sysds.runtime.controlprogram.federated.FederatedWorker;\nimport org.apache.sysds.runtime.controlprogram.parfor.stat.InfrastructureAnalyzer;\nimport org.apache.sysds.runtime.controlprogram.parfor.util.IDHandler;\nimport org.apache.sysds.runtime.instructions.gpu.context.GPUContextPool;\n+import org.apache.sysds.runtime.instructions.spark.utils.SparkUtils;\nimport org.apache.sysds.runtime.io.IOUtilFunctions;\nimport org.apache.sysds.runtime.lineage.LineageCacheConfig;\nimport org.apache.sysds.runtime.lineage.LineageCacheConfig.LineageCachePolicy;\n@@ -515,6 +516,9 @@ public class DMLScript\n//0) cleanup federated workers if necessary\nFederatedData.clearFederatedWorkers();\n+ //0) shutdown prefetch/broadcast thread pool if necessary\n+ SparkUtils.shutdownPool();\n+\n//1) cleanup scratch space (everything for current uuid)\n//(required otherwise export to hdfs would skip assumed unnecessary writes if same name)\nHDFSTool.deleteFileIfExistOnHDFS( config.getTextValue(DMLConfig.SCRATCH_SPACE) + dirSuffix );\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/lops/compile/Dag.java", "new_path": "src/main/java/org/apache/sysds/lops/compile/Dag.java", "diff": "@@ -233,6 +233,7 @@ public class Dag<N extends Lop>\nfor (Lop l : nodes) {\nnodesWithPrefetch.add(l);\nif (isPrefetchNeeded(l)) {\n+ //TODO: No prefetch if the parent is placed right after the spark OP\nList<Lop> oldOuts = new ArrayList<>(l.getOutputs());\n//Construct a Prefetch lop that takes this Spark node as a input\nUnaryCP prefetch = new UnaryCP(l, OpOp1.PREFETCH, l.getDataType(), l.getValueType(), ExecType.CP);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/PrefetchCPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/PrefetchCPInstruction.java", "diff": "@@ -49,8 +49,8 @@ public class PrefetchCPInstruction extends UnaryCPInstruction {\n// If the next instruction which takes this output as an input comes before\n// the prefetch thread triggers, that instruction will start the operations.\n// In that case this Prefetch instruction will act like a NOOP.\n- if (SparkUtils.triggerRDDThread == null)\n- SparkUtils.triggerRDDThread = Executors.newSingleThreadExecutor();\n- SparkUtils.triggerRDDThread.submit(new TriggerRDDOperationsTask(ec.getMatrixObject(output)));\n+ if (SparkUtils.triggerRDDPool == null)\n+ SparkUtils.triggerRDDPool = Executors.newCachedThreadPool();\n+ SparkUtils.triggerRDDPool.submit(new TriggerRDDOperationsTask(ec.getMatrixObject(output)));\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/utils/SparkUtils.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/utils/SparkUtils.java", "diff": "@@ -63,7 +63,7 @@ import java.util.stream.LongStream;\npublic class SparkUtils\n{\n- public static ExecutorService triggerRDDThread = null;\n+ public static ExecutorService triggerRDDPool = null;\n//internal configuration\npublic static final StorageLevel DEFAULT_TMP = Checkpoint.DEFAULT_STORAGE_LEVEL;\n@@ -296,6 +296,14 @@ public class SparkUtils\nmo.acquireReadAndRelease();\n}\n+ public static void shutdownPool() {\n+ if (triggerRDDPool != null) {\n+ //shutdown prefetch/broadcast thread pool\n+ triggerRDDPool.shutdown();\n+ triggerRDDPool = null;\n+ }\n+ }\n+\nprivate static class CheckSparsityFunction implements VoidFunction<Tuple2<MatrixIndexes,MatrixBlock>>\n{\nprivate static final long serialVersionUID = 4150132775681848807L;\n" } ]

[ { "change_type": "ADD", "old_path": null, "new_path": "scripts/builtin/selectByVarThresh.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+m_selectByVarThresh = function(Matrix[Double] X, Double thresh = 0)\n+ return (Matrix[Double] Xp, Matrix[Double] I)\n+{\n+ # drop feature with <= thresh variance, by default drop constants\n+ I = (colVars(X) > thresh);\n+ Xp = removeEmpty(target=X, margin=\"cols\", select=I);\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/common/Builtins.java", "new_path": "src/main/java/org/apache/sysds/common/Builtins.java", "diff": "@@ -232,6 +232,7 @@ public enum Builtins {\nROWVAR(\"rowVars\", false),\nSAMPLE(\"sample\", false),\nSD(\"sd\", false),\n+ SELVARTHRESH(\"selectByVarThresh\", true),\nSEQ(\"seq\", false),\nSHERLOCK(\"sherlock\", true),\nSHERLOCKPREDICT(\"sherlockPredict\", true),\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/parser/dml/DmlSyntacticValidator.java", "new_path": "src/main/java/org/apache/sysds/parser/dml/DmlSyntacticValidator.java", "diff": "@@ -558,11 +558,12 @@ public class DmlSyntacticValidator implements DmlListener {\n}\n@Override\n- public void exitFunctionCallMultiAssignmentStatement(\n- FunctionCallMultiAssignmentStatementContext ctx) {\n+ public void exitFunctionCallMultiAssignmentStatement(FunctionCallMultiAssignmentStatementContext ctx) {\n+ if( ctx.name == null )\n+ throw new ParseException(\"Missing name of multi-assignment function call (see parser issues above).\");\nString[] names = getQualifiedNames(ctx.name.getText());\nif(names == null) {\n- notifyErrorListeners(\"incorrect function name (only namespace.functionName allowed. Hint: If you are trying to use builtin functions, you can skip the namespace)\", ctx.name);\n+ notifyErrorListeners(\"incorrect function name (only namespace::functionName allowed. Hint: If you are trying to use builtin functions, you can skip the namespace)\", ctx.name);\nreturn;\n}\nString namespace = names[0];\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/EvalNaryCPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/EvalNaryCPInstruction.java", "diff": "@@ -159,6 +159,10 @@ public class EvalNaryCPInstruction extends BuiltinNaryCPInstruction {\nmb = DataConverter.convertToMatrixBlock(((FrameObject) newOutput).acquireRead());\nec.cleanupCacheableData((FrameObject) newOutput);\n}\n+ else {\n+ throw new DMLRuntimeException(\"Invalid eval return type: \"+newOutput.getDataType().name()\n+ + \" (valid: matrix/frame/scalar; where frames or scalars are converted to output matrices)\");\n+ }\noutputMO.acquireModify(mb);\noutputMO.release();\nec.setVariable(output.getName(), outputMO);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/CtableFEDInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/fed/CtableFEDInstruction.java", "diff": "@@ -197,7 +197,7 @@ public class CtableFEDInstruction extends ComputationFEDInstruction {\ncontinue;\n// check intersect with AND and compare number of nnz\n- MatrixBlock prevExtend = new MatrixBlock(curr.getNumRows(), curr.getNumColumns(), 0.0);\n+ MatrixBlock prevExtend = new MatrixBlock(curr.getNumRows(), curr.getNumColumns(), true, 0);\nprevExtend.copy(0, prev.getNumRows()-1, 0, prev.getNumColumns()-1, prev, true);\nMatrixBlock intersect = curr.binaryOperationsInPlace(new BinaryOperator(And.getAndFnObject()), prevExtend);\n@@ -240,7 +240,7 @@ public class CtableFEDInstruction extends ComputationFEDInstruction {\n}\nprivate static MatrixBlock aggResult(Future<FederatedResponse>[] ffr) {\n- MatrixBlock resultBlock = new MatrixBlock(1, 1, 0);\n+ MatrixBlock resultBlock = new MatrixBlock(1, 1, true, 0);\nint dim1 = 0, dim2 = 0;\nfor(int i = 0; i < ffr.length; i++) {\ntry {\n@@ -249,10 +249,10 @@ public class CtableFEDInstruction extends ComputationFEDInstruction {\ndim2 = mb.getNumColumns() > dim2 ? mb.getNumColumns() : dim2;\n// set next and prev to same output dimensions\n- MatrixBlock prev = new MatrixBlock(dim1, dim2, 0.0);\n+ MatrixBlock prev = new MatrixBlock(dim1, dim2, true, 0);\nprev.copy(0, resultBlock.getNumRows()-1, 0, resultBlock.getNumColumns()-1, resultBlock, true);\n- MatrixBlock next = new MatrixBlock(dim1, dim2, 0.0);\n+ MatrixBlock next = new MatrixBlock(dim1, dim2, true, 0);\nnext.copy(0, mb.getNumRows()-1, 0, mb.getNumColumns()-1, mb, true);\n// add worker results\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/docs/source/guide/federated.rst", "new_path": "src/main/python/docs/source/guide/federated.rst", "diff": "@@ -82,7 +82,7 @@ The aggregated sum using federated instructions in python SystemDS is done as fo\naddress = \"localhost:8001/temp/test.csv\"\nwith SystemDSContext() as sds:\n- fed_a = sds.federated(sds, [address], [dims])\n+ fed_a = sds.federated([address], [dims])\n# Sum the federated matrix and call compute to execute\nprint(fed_a.sum().compute())\n# Result should be 45.\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/federated/FederatedWorkerHandler.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/federated/FederatedWorkerHandler.java", "diff": "@@ -197,6 +197,7 @@ public class FederatedWorkerHandler extends ChannelInboundHandlerAdapter {\nString delim = null;\nFileSystem fs = null;\nMetaDataAll mtd;\n+\ntry {\nString mtdname = DataExpression.getMTDFileName(filename);\nPath path = new Path(mtdname);\n@@ -219,9 +220,7 @@ public class FederatedWorkerHandler extends ChannelInboundHandlerAdapter {\nthrow ex;\n}\ncatch (Exception ex) {\n- String msg = \"Exception in reading metadata of: \" + filename;\n- log.error(msg, ex);\n- throw new DMLRuntimeException(msg);\n+ throw new DMLRuntimeException(ex);\n}\nfinally {\nIOUtilFunctions.closeSilently(fs);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/meta/MetaDataAll.java", "new_path": "src/main/java/org/apache/sysds/runtime/meta/MetaDataAll.java", "diff": "@@ -39,6 +39,7 @@ import org.apache.sysds.parser.Expression;\nimport org.apache.sysds.parser.LanguageException;\nimport org.apache.sysds.parser.ParseException;\nimport org.apache.sysds.parser.StringIdentifier;\n+import org.apache.sysds.runtime.DMLRuntimeException;\nimport org.apache.sysds.runtime.controlprogram.caching.CacheableData;\nimport org.apache.sysds.runtime.io.IOUtilFunctions;\nimport org.apache.sysds.runtime.privacy.PrivacyConstraint;\n@@ -50,6 +51,7 @@ import org.apache.wink.json4j.JSONException;\nimport org.apache.wink.json4j.JSONObject;\npublic class MetaDataAll extends DataIdentifier {\n+ // private static final Log LOG = LogFactory.getLog(MetaDataAll.class.getName());\nprivate JSONObject _metaObj;\n@@ -79,8 +81,8 @@ public class MetaDataAll extends DataIdentifier {\ntry {\n_metaObj = JSONHelper.parse(br);\n}\n- catch(IOException e) {\n- e.printStackTrace();\n+ catch(Exception e) {\n+ throw new DMLRuntimeException(e);\n}\nsetPrivacy(PrivacyConstraint.PrivacyLevel.None);\nparseMetaDataParams();\n@@ -174,7 +176,14 @@ public class MetaDataAll extends DataIdentifier {\ncase DataExpression.FINE_GRAINED_PRIVACY: setFineGrainedPrivacy(val.toString()); break;\ncase DataExpression.DELIM_DELIMITER: setDelim(val.toString()); break;\ncase DataExpression.SCHEMAPARAM: setSchema(val.toString()); break;\n- case DataExpression.DELIM_HAS_HEADER_ROW: setHasHeader(true);\n+ case DataExpression.DELIM_HAS_HEADER_ROW:\n+ if(val instanceof Boolean){\n+ boolean valB = (Boolean) val;\n+ setHasHeader(valB);\n+ break;\n+ }\n+ else\n+ setHasHeader(false);\ncase DataExpression.DELIM_SPARSE: setSparseDelim((boolean) val);\n}\n}\n@@ -402,4 +411,9 @@ public class MetaDataAll extends DataIdentifier {\n}\nreturn false;\n}\n+\n+ @Override\n+ public String toString() {\n+ return \"MetaDataAll\\n\" + _metaObj + \"\\n\" + super.toString();\n+ }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/python/tests/federated/runFedTest.sh", "new_path": "src/main/python/tests/federated/runFedTest.sh", "diff": "workerdir=\"tests/federated/worker/\"\noutputdir=\"tests/federated/output/\"\ntmpfiledir=\"tests/federated/tmp/\"\n-mkdir $workerdir\n-mkdir $outputdir\n+mkdir -p $workerdir\n+mkdir -p $outputdir\nw1_Output=\"$workerdir/w1\"\nw2_Output=\"$workerdir/w2\"\nlog=\"$outputdir/out.log\"\n@@ -55,13 +55,16 @@ echo -e \"\\nWorker 1:\"\ncat $w1_Output\necho -e \"\\nWorker 2:\"\ncat $w2_Output\n-rm -r $workerdir\necho -e \"\\n------------\\nTest output:\\n------------\"\ncat $log\ngrepvals=\"$(tail -n 10 $log | grep OK)\"\n+echo -e \"------------\\n\"\n+\n+# Cleanup\n+rm -r $workerdir\nrm -r $outputdir\nrm -r $tmpfiledir\n-echo -e \"------------\\n\"\n+\nif [[ $grepvals == *\"OK\"* ]]; then\nexit 0\nelse\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/python/tests/federated/test_federated_aggregations_noHeader.py", "diff": "+# -------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+# -------------------------------------------------------------\n+\n+import io\n+import json\n+import os\n+import shutil\n+import sys\n+import unittest\n+\n+import numpy as np\n+from systemds.context import SystemDSContext\n+\n+os.environ['SYSDS_QUIET'] = \"1\"\n+\n+dim = 3\n+\n+m1 = np.asarray([[1, 2, 3], [4, 5, 6], [7, 8, 9]], dtype=np.int16)\n+m2 = np.asarray([[2, 2, 2], [3, 3, 3], [4, 4, 4]], dtype=np.int16)\n+\n+tempdir = \"./tests/federated/tmp/test_federated_aggregations_noHeader/\"\n+mtd = {\"format\": \"csv\", \"header\": False, \"rows\": dim,\n+ \"cols\": dim, \"data_type\": \"matrix\", \"value_type\": \"double\"}\n+\n+# Create the testing directory if it does not exist.\n+if not os.path.exists(tempdir):\n+ os.makedirs(tempdir)\n+\n+# Save data files for the Federated workers.\n+np.savetxt(tempdir + \"m1.csv\", m1, delimiter=\",\",fmt='%d')\n+with io.open(tempdir + \"m1.csv.mtd\", \"w\", encoding=\"utf-8\") as f:\n+ f.write(json.dumps(mtd, ensure_ascii=False))\n+\n+np.savetxt(tempdir + \"m2.csv\", m2, delimiter=\",\",fmt='%d')\n+with io.open(tempdir + \"m2.csv.mtd\", \"w\", encoding=\"utf-8\") as f:\n+ f.write(json.dumps(mtd, ensure_ascii=False))\n+\n+# Federated workers + file locations\n+fed1 = \"localhost:8001/\" + tempdir + \"m1.csv\"\n+fed2 = \"localhost:8002/\" + tempdir + \"m2.csv\"\n+\n+\n+class TestFederatedAggFn(unittest.TestCase):\n+\n+ sds: SystemDSContext = None\n+\n+ @classmethod\n+ def setUpClass(cls):\n+ cls.sds = SystemDSContext()\n+\n+ @classmethod\n+ def tearDownClass(cls):\n+ cls.sds.close()\n+\n+ def test_equals(self):\n+ f_m = (\n+ self.sds.federated(\n+ [fed1],\n+ [([0, 0], [dim, dim])])\n+ .compute()\n+ )\n+ self.assertTrue(np.allclose(f_m, m1))\n+\n+ def test_sum3(self):\n+ # [[m1,m1,m1,m1,m1,m2,m2,m2,m2,m2]\n+ # [m1,m1,m1,m1,m1,m2,m2,m2,m2,m2]\n+ # [m1,m1,m1,m1,m1,m2,m2,m2,m2,m2]\n+ # [m1,m1,m1,m1,m1,m2,m2,m2,m2,m2]\n+ # [m1,m1,m1,m1,m1,m2,m2,m2,m2,m2]]\n+ f_m_a = (\n+ self.sds.federated(\n+ [fed1, fed2],\n+ [([0, 0], [dim, dim]), ([0, dim], [dim, dim * 2])])\n+ .sum()\n+ .compute()\n+ )\n+ m1_m2 = m1.sum() + m2.sum()\n+ self.assertAlmostEqual(f_m_a, m1_m2)\n+\n+ def test_sum1(self):\n+ f_m1 = (\n+ self.sds.federated(\n+ [fed1],\n+ [([0, 0], [dim, dim])])\n+ .sum()\n+ .compute()\n+ )\n+ m1_r = m1.sum()\n+ self.assertAlmostEqual(f_m1, m1_r)\n+\n+ def test_sum2(self):\n+ f_m2 = (\n+ self.sds.federated(\n+ [fed2],\n+ [([0, 0], [dim, dim])])\n+ .sum()\n+ .compute()\n+ )\n+ m2_r = m2.sum()\n+ self.assertAlmostEqual(f_m2, m2_r)\n+\n+ def test_sum3(self):\n+ # [[m1,m1,m1,m1,m1,m2,m2,m2,m2,m2]\n+ # [m1,m1,m1,m1,m1,m2,m2,m2,m2,m2]\n+ # [m1,m1,m1,m1,m1,m2,m2,m2,m2,m2]\n+ # [m1,m1,m1,m1,m1,m2,m2,m2,m2,m2]\n+ # [m1,m1,m1,m1,m1,m2,m2,m2,m2,m2]]\n+ f_m1_m2 = (\n+ self.sds.federated(\n+ [fed1, fed2],\n+ [([0, 0], [dim, dim]), ([0, dim], [dim, dim * 2])])\n+ .sum()\n+ .compute()\n+ )\n+\n+ m1_m2 = np.concatenate((m1, m2), axis=1).sum()\n+\n+ self.assertAlmostEqual(f_m1_m2, m1_m2)\n+\n+ def test_sum4(self):\n+ # [[m1,m1,m1,m1,m1]\n+ # [m1,m1,m1,m1,m1]\n+ # [m1,m1,m1,m1,m1]\n+ # [m1,m1,m1,m1,m1]\n+ # [m1,m1,m1,m1,m1]\n+ # [m2,m2,m2,m2,m2]\n+ # [m2,m2,m2,m2,m2]\n+ # [m2,m2,m2,m2,m2]\n+ # [m2,m2,m2,m2,m2]\n+ # [m2,m2,m2,m2,m2]]\n+ f_m1_m2 = (\n+ self.sds.federated(\n+ [fed1, fed2],\n+ [([0, 0], [dim, dim]), ([dim, 0], [dim * 2, dim])])\n+ .sum()\n+ .compute()\n+ )\n+ m1_m2 = np.concatenate((m1, m2)).sum()\n+ self.assertAlmostEqual(f_m1_m2, m1_m2)\n+\n+ # -----------------------------------\n+ # The rest of the tests are\n+ # Extended functionality not working Yet\n+ # -----------------------------------\n+\n+ def test_sum5(self):\n+ # [[m1,m1,m1,m1,m1, 0, 0, 0, 0, 0]\n+ # [m1,m1,m1,m1,m1, 0, 0, 0, 0, 0]\n+ # [m1,m1,m1,m1,m1,m2,m2,m2,m2,m2]\n+ # [m1,m1,m1,m1,m1,m2,m2,m2,m2,m2]\n+ # [m1,m1,m1,m1,m1,m2,m2,m2,m2,m2]\n+ # [ 0, 0, 0, 0, 0,m2,m2,m2,m2,m2]\n+ # [ 0, 0, 0, 0, 0,m2,m2,m2,m2,m2]]\n+ f_m_a = (\n+ self.sds.federated(\n+ [fed1, fed2],\n+ [([0, 0], [dim, dim]), ([2, dim], [dim + 2, dim * 2])])\n+ .sum()\n+ .compute()\n+ )\n+ m1_m2 = m1.sum() + m2.sum()\n+ self.assertAlmostEqual(f_m_a, m1_m2)\n+\n+ def test_sum8(self):\n+ # [[ 0, 0, 0, 0, 0, 0, 0, 0]\n+ # [ 0, 0, 0, 0, 0, 0, 0, 0]\n+ # [ 0, 0, 0,m1,m1,m1,m1,m1]\n+ # [ 0, 0, 0,m1,m1,m1,m1,m1]\n+ # [ 0, 0, 0,m1,m1,m1,m1,m1]\n+ # [ 0, 0, 0,m1,m1,m1,m1,m1]\n+ # [ 0, 0, 0,m1,m1,m1,m1,m1]]\n+ f_m_a = (\n+ self.sds.federated(\n+ [fed1],\n+ [([2, 3], [dim + 2, dim + 3])])\n+ .sum()\n+ .compute()\n+ )\n+\n+ m = m1.sum()\n+\n+ self.assertAlmostEqual(f_m_a, m)\n+\n+\n+if __name__ == \"__main__\":\n+ unittest.main(exit=False)\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/AutomatedTestBase.java", "new_path": "src/test/java/org/apache/sysds/test/AutomatedTestBase.java", "diff": "@@ -69,6 +69,7 @@ import org.apache.sysds.runtime.controlprogram.federated.FederatedData;\nimport org.apache.sysds.runtime.controlprogram.federated.FederatedRange;\nimport org.apache.sysds.runtime.controlprogram.federated.FederationMap;\nimport org.apache.sysds.runtime.controlprogram.federated.FederationUtils;\n+import org.apache.sysds.runtime.io.FileFormatProperties;\nimport org.apache.sysds.runtime.io.FileFormatPropertiesCSV;\nimport org.apache.sysds.runtime.io.FrameReader;\nimport org.apache.sysds.runtime.io.FrameReaderFactory;\n@@ -515,13 +516,7 @@ public abstract class AutomatedTestBase {\nString completePath = baseDirectory + INPUT_DIR + name + \"/in\";\nString completeRPath = baseDirectory + INPUT_DIR + name + \".mtx\";\n- try {\n- cleanupExistingData(baseDirectory + INPUT_DIR + name, bIncludeR);\n- }\n- catch(IOException e) {\n- e.printStackTrace();\n- throw new RuntimeException(e);\n- }\n+ cleanupDir(baseDirectory + INPUT_DIR + name, bIncludeR);\nTestUtils.writeTestMatrix(completePath, matrix);\nif(bIncludeR) {\n@@ -535,6 +530,30 @@ public abstract class AutomatedTestBase {\nreturn matrix;\n}\n+ protected void writeCSVMatrix(String name, double[][] matrix, boolean header, MatrixCharacteristics mc) {\n+ try {\n+ final String completePath = baseDirectory + INPUT_DIR + name;\n+ final String completeMTDPath = baseDirectory + INPUT_DIR + name + \".mtd\";\n+ cleanupDir(completePath, false);\n+ TestUtils.writeCSV(completePath, matrix, header);\n+ final FileFormatProperties ffp = header ? new FileFormatPropertiesCSV(true, \",\", false, 0.0, \"\") : new FileFormatPropertiesCSV();\n+ HDFSTool.writeMetaDataFile(completeMTDPath, ValueType.FP64, mc, FileFormat.CSV, ffp);\n+ }\n+ catch(Exception e) {\n+ throw new RuntimeException(e);\n+ }\n+ }\n+\n+ protected void cleanupDir(String fullPath, boolean bIncludeR){\n+ try {\n+ cleanupExistingData(fullPath, bIncludeR);\n+ }\n+ catch(IOException e) {\n+ e.printStackTrace();\n+ throw new RuntimeException(e);\n+ }\n+ }\n+\nprotected double[][] writeInputMatrixWithMTD(String name, MatrixBlock matrix, boolean bIncludeR) {\ndouble[][] data = DataConverter.convertToDoubleMatrix(matrix);\nreturn writeInputMatrixWithMTD(name, data, bIncludeR);\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/TestUtils.java", "new_path": "src/test/java/org/apache/sysds/test/TestUtils.java", "diff": "@@ -64,6 +64,7 @@ import org.apache.hadoop.io.SequenceFile;\nimport org.apache.hadoop.io.SequenceFile.Writer;\nimport org.apache.sysds.common.Types.FileFormat;\nimport org.apache.sysds.common.Types.ValueType;\n+import org.apache.sysds.runtime.DMLRuntimeException;\nimport org.apache.sysds.runtime.data.TensorBlock;\nimport org.apache.sysds.runtime.io.FrameWriter;\nimport org.apache.sysds.runtime.io.FrameWriterFactory;\n@@ -2040,6 +2041,31 @@ public class TestUtils\n}\n}\n+\n+ protected static void writeCSV(String completePath, double[][] matrix, boolean header) throws IOException{\n+ Path path = new Path(completePath);\n+ FileSystem fs = IOUtilFunctions.getFileSystem(path, conf);\n+ DataOutputStream out = fs.create(path, true);\n+ try(BufferedWriter pw = new BufferedWriter(new OutputStreamWriter(out))) {\n+\n+ if(header) {\n+ pw.append(\"d0\");\n+ for(int i = 1; i < matrix[0].length; i++) {\n+ pw.append(\",d\" + i);\n+ }\n+ pw.append(\"\\n\");\n+ }\n+ for(int j = 0; j < matrix.length; j++) {\n+ pw.append(\"\" + matrix[j][0]);\n+ for(int i = 1; i < matrix[j].length; i++) {\n+ pw.append(\",\" + matrix[j][i]);\n+ }\n+ pw.append(\"\\n\");\n+ }\n+ }\n+ }\n+\n+\n/**\n* <p>\n* Writes a matrix to a file using the text format.\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysds/test/functions/federated/io/FederatedReaderCSV.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+package org.apache.sysds.test.functions.federated.io;\n+\n+import org.apache.commons.logging.Log;\n+import org.apache.commons.logging.LogFactory;\n+import org.apache.sysds.common.Types;\n+import org.apache.sysds.common.Types.ExecType;\n+import org.apache.sysds.runtime.controlprogram.caching.MatrixObject;\n+import org.apache.sysds.runtime.meta.MatrixCharacteristics;\n+import org.apache.sysds.test.AutomatedTestBase;\n+import org.apache.sysds.test.TestConfiguration;\n+import org.apache.sysds.test.TestUtils;\n+import org.apache.sysds.test.functions.federated.FederatedTestObjectConstructor;\n+import org.junit.Assert;\n+import org.junit.Test;\n+\[email protected]\n+public class FederatedReaderCSV extends AutomatedTestBase {\n+\n+ private static final Log LOG = LogFactory.getLog(FederatedReaderCSV.class.getName());\n+ private final static String TEST_DIR = \"functions/federated/ioR/\";\n+ private final static String TEST_NAME = \"FederatedReaderTest\";\n+ private final static String TEST_CLASS_DIR = TEST_DIR + FederatedReaderCSV.class.getSimpleName() + \"/\";\n+ private final static int blocksize = 1024;\n+\n+ private final static int dim = 3;\n+ long[][] begins = new long[][] {new long[] {0, 0}};\n+ long[][] ends = new long[][] {new long[] {dim, dim}};\n+\n+ @Override\n+ public void setUp() {\n+ TestUtils.clearAssertionInformation();\n+ addTestConfiguration(TEST_NAME, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME,new String[] {\"X1\"}));\n+ }\n+\n+ @Test\n+ public void testWithHeader() {\n+ federatedRead(true);\n+ }\n+\n+ @Test\n+ public void testWithoutHeader() {\n+ federatedRead(false);\n+ }\n+\n+ public void federatedRead( boolean header) {\n+ Types.ExecMode oldPlatform = setExecMode(ExecType.CP);\n+ getAndLoadTestConfiguration(TEST_NAME);\n+ setOutputBuffering(true);\n+\n+\n+ // empty script name because we don't execute any script, just start the worker\n+\n+ fullDMLScriptName = \"\";\n+ int port1 = getRandomAvailablePort();\n+ Thread t1 = startLocalFedWorkerThread(port1);\n+ String host = \"localhost\";\n+\n+ try {\n+ double[][] X1 = new double[][] {new double[] {1, 2, 3}, new double[] {4, 5, 6}, new double[] {7, 8, 9}};\n+ MatrixCharacteristics mc = new MatrixCharacteristics(dim, dim, blocksize, dim * dim);\n+ writeCSVMatrix(\"X1\", X1, header, mc);\n+\n+ // Thread.sleep(10000);\n+ MatrixObject fed = FederatedTestObjectConstructor.constructFederatedInput(dim, dim, blocksize, host, begins,\n+ ends, new int[] {port1}, new String[] {input(\"X1\")}, input(\"X.json\"));\n+ writeInputFederatedWithMTD(\"X.json\", fed, null);\n+\n+ // Run reference dml script with normal matrix\n+\n+ fullDMLScriptName = SCRIPT_DIR + \"functions/federated/io/\" + TEST_NAME + \"1Reference.dml\";\n+ programArgs = new String[] {\"-stats\", \"-args\", input(\"X1\")};\n+\n+ String refOut = runTest(null).toString();\n+\n+ LOG.debug(refOut);\n+\n+ // Run federated\n+ fullDMLScriptName = SCRIPT_DIR + \"functions/federated/io/\" + TEST_NAME + \".dml\";\n+ programArgs = new String[] {\"-stats\", \"-args\", input(\"X.json\")};\n+ String out = runTest(null).toString();\n+\n+ Assert.assertTrue(heavyHittersContainsString(\"fed_uak+\"));\n+ // Verify output\n+ Assert.assertEquals(Double.parseDouble(refOut.split(\"\\n\")[0]), Double.parseDouble(out.split(\"\\n\")[0]),\n+ 0.00001);\n+ }\n+ catch(Exception e) {\n+ e.printStackTrace();\n+ Assert.assertTrue(false);\n+ }\n+ finally {\n+ resetExecMode(oldPlatform);\n+ }\n+\n+ TestUtils.shutdownThreads(t1);\n+ }\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/federated/io/FederatedReaderTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/federated/io/FederatedReaderTest.java", "diff": "@@ -67,13 +67,11 @@ public class FederatedReaderTest extends AutomatedTestBase {\n@Test\npublic void federatedSingleNodeReadOneWorker() {\n- LOG.debug(\"1Federated\");\nfederatedRead(Types.ExecMode.SINGLE_NODE, 1);\n}\n@Test\npublic void federatedSingleNodeReadTwoWorker() {\n- LOG.debug(\"2Federated\");\nfederatedRead(Types.ExecMode.SINGLE_NODE, 2);\n}\n@@ -124,6 +122,8 @@ public class FederatedReaderTest extends AutomatedTestBase {\nString refOut = runTest(null).toString();\n+ LOG.debug(refOut);\n+\n// Run federated\nfullDMLScriptName = SCRIPT_DIR + \"functions/federated/io/\" + TEST_NAME + \".dml\";\nprogramArgs = new String[] {\"-stats\", \"-args\", input(\"X.json\")};\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/TestUtils.java", "new_path": "src/test/java/org/apache/sysds/test/TestUtils.java", "diff": "@@ -64,7 +64,6 @@ import org.apache.hadoop.io.SequenceFile;\nimport org.apache.hadoop.io.SequenceFile.Writer;\nimport org.apache.sysds.common.Types.FileFormat;\nimport org.apache.sysds.common.Types.ValueType;\n-import org.apache.sysds.runtime.DMLRuntimeException;\nimport org.apache.sysds.runtime.data.TensorBlock;\nimport org.apache.sysds.runtime.io.FrameWriter;\nimport org.apache.sysds.runtime.io.FrameWriterFactory;\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/federated/primitives/FederatedTokenizeTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/federated/primitives/FederatedTokenizeTest.java", "diff": "@@ -125,7 +125,7 @@ public class FederatedTokenizeTest extends AutomatedTestBase {\n// Run reference dml script with normal matrix\nfullDMLScriptName = HOME + TEST_NAME + \"Reference.dml\";\n- programArgs = new String[] {\"-explain\", \"-args\", DATASET_DIR + DATASET, HOME + TEST_NAME + \".json\", expected(\"S\")};\n+ programArgs = new String[] {\"-explain\", \"-args\", input(\"AH\"), HOME + TEST_NAME + \".json\", expected(\"S\")};\nrunTest(null);\n// Run actual dml script with federated matrix\n@@ -137,7 +137,12 @@ public class FederatedTokenizeTest extends AutomatedTestBase {\n\"in_S=\" + input(HOME + TEST_NAME + \".json\"), \"rows=\" + rows, \"cols=\" + cols,\n\"out_R=\" + output(\"S\")};\nrunTest(null);\n-\n+ try{\n+ Thread.sleep(10000);\n+ }\n+ catch(Exception e){\n+ throw new RuntimeException(e);\n+ }\ncompareResults(1e-9);\nAssert.assertTrue(heavyHittersContainsString(\"fed_tokenize\"));\nTestUtils.shutdownThreads(t1, t2, t3, t4);\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/federated/FederatedTokenizeTest.dml", "new_path": "src/test/scripts/functions/federated/FederatedTokenizeTest.dml", "diff": "#-------------------------------------------------------------\nF1 = federated(type=\"frame\", addresses=list($in_X1, $in_X2, $in_X3),\n- ranges=list(list(0, 0), list(2, $cols), list(2, 0), list(4, $cols),\n- list(4, 0), list(6, $cols)));\n+ ranges=list(list(0, 0), list(3, $cols), list(3, 0), list(6, $cols),\n+ list(6, 0), list(9, $cols)));\nmax_token = 2000;\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/federated/FederatedTokenizeTestReference.dml", "new_path": "src/test/scripts/functions/federated/FederatedTokenizeTestReference.dml", "diff": "#\n#-------------------------------------------------------------\n-F = read($1, data_type=\"frame\", format=\"csv\", sep=\",\");\n-F = F[2:3, 1:4];\n+F = read($1);\nF = rbind(F, rbind(F, F));\nmax_token = 2000;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/federated/FederatedTokenizeTest.dml", "new_path": "src/test/scripts/functions/federated/FederatedTokenizeTest.dml", "diff": "#-------------------------------------------------------------\nF1 = federated(type=\"frame\", addresses=list($in_X1, $in_X2, $in_X3),\n- ranges=list(list(0, 0), list(2, $cols), list(2, 0), list(4, $cols),\n- list(4, 0), list(6, $cols)));\n+ ranges=list(list(0, 0), list(3, $cols), list(3, 0), list(6, $cols),\n+ list(6, 0), list(9, $cols)));\nmax_token = 2000;\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/federated/FederatedTokenizeTestReference.dml", "new_path": "src/test/scripts/functions/federated/FederatedTokenizeTestReference.dml", "diff": "#\n#-------------------------------------------------------------\n-F = read($1, data_type=\"frame\", format=\"csv\", sep=\",\");\n-F = F[2:3, 1:4];\n+F = read($1);\nF = rbind(F, rbind(F, F));\nmax_token = 2000;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/paramserv/FederatedPSControlThread.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/paramserv/FederatedPSControlThread.java", "diff": "@@ -522,7 +522,9 @@ public class FederatedPSControlThread extends PSWorker implements Callable<Void>\nParamservUtils.accrueGradients(accGradients, gradients, false);\n// update the local model with gradients if needed\n- if((_localUpdate && batchCounter < _numBatchesToCompute - 1) | modelAvg) {\n+ // FIXME ensure that with modelAvg we always update the model\n+ // (current fails due to missing aggregation instruction)\n+ if(_localUpdate && batchCounter < _numBatchesToCompute - 1) {\n// Invoke the aggregate function\nassert aggregationInstruction != null;\naggregationInstruction.processInstruction(ec);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinExecutePipelineTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinExecutePipelineTest.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\npackage org.apache.sysds.test.functions.pipelines;\nimport org.apache.sysds.common.Types;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinExecutePipelineTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinExecutePipelineTest.java", "diff": "@@ -65,10 +65,4 @@ public class BuiltinExecutePipelineTest extends AutomatedTestBase {\nresetExecMode(modeOld);\n}\n}\n-\n-\n- public static void main(String[] args) {\n- String s = null;\n- System.out.println(\"length is \"+s.length());\n- }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkEvaluateTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkEvaluateTest.java", "diff": "@@ -24,7 +24,6 @@ import org.apache.sysds.test.AutomatedTestBase;\nimport org.apache.sysds.test.TestConfiguration;\nimport org.apache.sysds.test.TestUtils;\nimport org.junit.Assert;\n-import org.junit.Ignore;\nimport org.junit.Test;\npublic class BuiltinTopkEvaluateTest extends AutomatedTestBase {\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/intermediates/classification/bestAcc.csv", "new_path": "src/test/scripts/functions/pipelines/intermediates/classification/bestAcc.csv", "diff": "-85.58558558558559\n+84.68468468468468\n82.88288288288288\n82.88288288288288\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/intermediates/classification/hp.csv", "new_path": "src/test/scripts/functions/pipelines/intermediates/classification/hp.csv", "diff": "-36.0,3.0,3.0,2.0,1.0,0,0,0,1.0,0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,1.0,0,2.0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n-36.0,3.0,7.0,1.0,1.0,0,0,0,1.0,0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,1.0,0,2.0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n+36.0,3.0,2.0,2.0,1.0,0,0,0,1.0,0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,1.0,0,2.0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n36.0,3.0,1.0,1.0,1.0,0,0,0,1.0,0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,1.0,0,2.0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n+36.0,3.0,7.0,2.0,1.0,0,0,0,1.0,0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,1.0,0,2.0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/intermediates/classification/lp.csv", "new_path": "src/test/scripts/functions/pipelines/intermediates/classification/lp.csv", "diff": "-ED,EC,CI,DUMMY\n+ED,MVI,CI,DUMMY\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/intermediates/classification/pip.csv", "new_path": "src/test/scripts/functions/pipelines/intermediates/classification/pip.csv", "diff": "outlierBySd,imputeByMedian,wtomeklink,dummycoding\n-outlierBySd,imputeByMedian,wtomeklink,dummycoding\noutlierBySd,imputeByMean,wtomeklink,dummycoding\n+outlierBySd,imputeByMedian,wtomeklink,dummycoding\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/topkcleaningRegressionTest.dml", "new_path": "src/test/scripts/functions/pipelines/topkcleaningRegressionTest.dml", "diff": "@@ -60,12 +60,22 @@ write(result, $O)\n# choice of parameters provided by API, X, Y, clone_X, evalFunHp (hyper-param), trainML (boolean for optimizing hp internally or passed by externally )\nevalRegression = function(Matrix[Double] X, Matrix[Double] Y, Matrix[Double] Xtest, Matrix[Double] Ytest, Matrix[Double] Xorig=as.matrix(0),\nMatrix[Double] evalFunHp, Boolean trainML = FALSE)\n-return(Matrix[Double] accuracy)\n+return(Matrix[Double] output)\n{\n+ if(trainML == 1)\n+ {\n+ # do the gridsearch for hyper-parameters\n+ params = list(\"icpt\",\"reg\", \"tol\", \"maxi\");\n+ paramRanges = list(seq(0,2),10^seq(0,-4), 10^seq(-6,-12), 10^seq(1,3));\n+ [B1, opt] = utils::topk_gridSearch(X=X, y=Y, train=\"lm\", predict=\"wmape\",\n+ numB=ncol(X)+1, cv=TRUE, params=params, paramValues=paramRanges, verbose=FALSE);\n+ evalFunHp = as.matrix(opt)\n+ }\nbeta = lm(X=X, y=Y, icpt=as.scalar(evalFunHp[1,1]), reg=as.scalar(evalFunHp[1,2]), tol=as.scalar(evalFunHp[1,3]),\nmaxi=as.scalar(evalFunHp[1,4]));\nacc = wmape(Xtest, Ytest, beta, as.scalar(evalFunHp[1,1]))\naccuracy = (1 - acc)\n+ output = cbind(accuracy, evalFunHp)\n}\nwmape = function(Matrix[Double] X, Matrix[Double] y, Matrix[Double] B, Integer icpt) return (Matrix[Double] loss) {\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/builtin/applyAndEvaluate.dml", "new_path": "scripts/builtin/applyAndEvaluate.dml", "diff": "@@ -49,7 +49,6 @@ return (Matrix[Double] result)\n}\n# # # when the evaluation function is called first we also compute and keep hyperparams of target application\ndirtyScore = getDirtyScore(X=Xtrain, Y=eYtrain, Xtest=Xtest, Ytest=eYtest, metaList=metaList, evaluationFunc=evaluationFunc, evalFunHp=evalFunHp)\n- print(\"dirty score: \"+dirtyScore)\n[Xtrain, Xtest] = runStringPipeline(Xtrain, Xtest, schema, mask, FALSE, correctTypos)\n# # # if mask has 1s then there are categorical features\n@@ -63,7 +62,6 @@ return (Matrix[Double] result)\nno_of_param = as.scalar(hp[1, 1]) + 1\nhp_width= hp[1, 2:no_of_param]\nhp_matrix = matrix(hp_width, rows=ncol(pip), cols=ncol(hp_width)/ncol(pip))\n- print(\"hp matrix:\\n\"+toString(hp_matrix))\npipList = list(lp = lp, ph = pip, hp = hp_matrix, flags = no_of_flag_vars)\n# argList = list(X=X, Y=Y, Xtest=Xtest, Ytest=Ytest, Xorig=clone_X, pipList=pipList, metaList=metaList, evalFunHp=evalFunHp, trainML=0)\n# # # now test accuracy\n" }, { "change_type": "MODIFY", "old_path": "scripts/builtin/bandit.dml", "new_path": "scripts/builtin/bandit.dml", "diff": "@@ -273,7 +273,6 @@ run_with_hyperparam = function(Frame[Unknown] lp, Frame[Unknown] ph_pip, Integer\npipList = list(lp = lp, ph = ph_pip[i], hp = hp_matrix, flags = no_of_flag_vars)\n[evalFunOutput, hpForPruning, changesByOp] = crossV(X=X, y=Y, cvk=cvk, evalFunHp=evalFunHp, pipList=pipList, metaList=metaList, hpForPruning=hpForPruning,\nchangesByOp=changesByOp, evalFunc=evaluationFunc, trainML = FALSE)\n- print(cvk+\" cross validations acc: \"+toString(evalFunOutput))\n}\nelse\n@@ -284,7 +283,6 @@ run_with_hyperparam = function(Frame[Unknown] lp, Frame[Unknown] ph_pip, Integer\nprint(\"Y contains only one class\")\nelse\nevalFunOutput = eval(evaluationFunc, list(X=eXtrain, Y=eYtrain, Xtest=eXtest, Ytest=eYtest, Xorig=as.matrix(0), evalFunHp=evalFunHp, trainML = 0))\n- print(\"holdout acc: \"+toString(evalFunOutput))\n}\n# evalFunOutput = eval(evaluationFunc, argList)\n@@ -646,7 +644,7 @@ return (Matrix[Double] accuracy, Matrix[Double] hpForPruning, Matrix[Double] cha\ntrainy = trainset[, 1]\ntestX = testset[, 2:ncol(testset)]\ntesty = testset[, 1]\n- # print(\"test in: \"+nrow(testy))\n+\nif(as.scalar(pipList['flags']) != 0)\n{\n[trainX, trainy, testX, testy, Tr, hpForPruning, changesByOp] = executePipeline(logical=as.frame(pipList['lp']), pipeline=as.frame(pipList['ph']),\n@@ -657,8 +655,6 @@ return (Matrix[Double] accuracy, Matrix[Double] hpForPruning, Matrix[Double] cha\nres = eval(evalFunc, list(X=trainX, Y=trainy, Xtest=testX, Ytest=testy, Xorig=as.matrix(0), evalFunHp=evalFunHp, trainML = 0))\naccuracyMatrix[i] = res\n}\n- print(cvk+\" CV: accuracy matrix: \\n\"+toString(accuracyMatrix))\n- print(cvk+\" CV: average accuracy: \"+mean(accuracyMatrix))\naccuracy = as.matrix(mean(accuracyMatrix))\n}\n@@ -674,7 +670,6 @@ return(Boolean execute)\n# get the non-zero index of hpForPruning\nidx = (hpForPruning > 0) * t(seq(1, ncol(hpForPruning)))\nidx = removeEmpty(target=idx, margin=\"cols\")\n- print(\"idx: \"+toString(idx))\nfor(i in 1:ncol(idx)) {\nindex = as.scalar(idx[1, i])\ninProcessHp = as.scalar(hp_matrix[index, 2])\n" }, { "change_type": "MODIFY", "old_path": "scripts/builtin/executePipeline.dml", "new_path": "scripts/builtin/executePipeline.dml", "diff": "@@ -87,11 +87,7 @@ s_executePipeline = function(Frame[String] logical = as.frame(\"NULL\"), Frame[Str\nif(as.scalar(pipeline[1, i]) == \"outlierBySd\" | as.scalar(pipeline[1, i]) == \"outlierByIQR\" | as.scalar(pipeline[1, i]) == \"imputeByFd\") {\nchanges = sum(abs(replace(target=Xout, pattern=NaN, replacement=0) - replace(target=as.matrix(hp[1]), pattern=NaN, replacement=0)) > 0.001 )\n[hpForPruning, changesByOp] = storeDataForPrunning(pipeline, hyperParameters, hpForPruning, changesByOp, changes, i)\n- print(\"ended \"+op+\" number of changes \"+changes)\n- # print(\"ended \"+op+\" number of changes \"+sum(abs(replace(target=X, pattern=NaN, replacement=0) - replace(target=Xclone, pattern=NaN, replacement=0)) > 0.001 ))\n}\n-\n- print(\"min max of Y: \"+min(Y)+\" \"+max(Y))\n}\nXtest = X[testStIdx:nrow(X), ]\nYtest = Y[testStIdx:nrow(X), ]\n@@ -298,7 +294,6 @@ return (Matrix[Double] X_filled)\n}\n}\nX_filled = X\n- print(\"imputeByFd: record changes: \"+sum(X_filled != X))\n}\n#######################################################################\n@@ -311,7 +306,6 @@ return (Matrix[Double] X_filled)\n{\noption = ifelse(op, \"locf\", \"nocb\")\nX_filled = na_locf(X=X, option=option, verbose=verbose)\n- print(\"nulls after forward_fill: \"+sum(is.na(X_filled)))\n}\n" }, { "change_type": "MODIFY", "old_path": "scripts/builtin/imputeByMean.dml", "new_path": "scripts/builtin/imputeByMean.dml", "diff": "@@ -58,5 +58,4 @@ return(Matrix[Double] X)\nq = table(seq(1, ncol(cX)), removeEmpty(target=seq(1, ncol(mask)), margin=\"rows\",\nselect=t(mask)), ncol(cX), ncol(X))\nX = (X_n %*% p) + (X_c %*% q)\n- print(\"imputeByMean: no of NaNs \"+sum(is.na(X)))\n}\n" }, { "change_type": "MODIFY", "old_path": "scripts/builtin/imputeByMedian.dml", "new_path": "scripts/builtin/imputeByMedian.dml", "diff": "@@ -62,5 +62,4 @@ return(Matrix[Double] X)\nq = table(seq(1, ncol(cX)), removeEmpty(target=seq(1, ncol(mask)), margin=\"rows\",\nselect=t(mask)), ncol(cX), ncol(X))\nX = (X_n %*% p) + (X_c %*% q)\n- print(\"imputeByMedian: no of NaNs \"+sum(is.na(X)))\n}\n\\ No newline at end of file\n" }, { "change_type": "MODIFY", "old_path": "scripts/pipelines/scripts/enumerateLogical.dml", "new_path": "scripts/pipelines/scripts/enumerateLogical.dml", "diff": "@@ -94,7 +94,6 @@ return (Frame[Unknown] bestLg, Double pre_best, Double T)\n# # sort the configurations groupwise\nmax_perf = bandit::getMaxPerConf(outPip, nrow(physicalConf))\nscores[i] = as.matrix(max_perf[1, 1])\n- print(\"scores: \\n\"+toString(scores))\n}\n# # select parents and best score\n@@ -119,8 +118,7 @@ return (Frame[Unknown] bestLg, Double pre_best, Double T)\n# # # if new best is not better than pre_best then no need od generating new population\nchildren = frame(0, rows=ceil(nrow(scores)/2), cols=pipLength)\ni = 1\n- print(i <= ceil(nrow(scores)/2))\n- print(converged)\n+\nwhile(i <= ceil(nrow(scores)/2) & !converged)\n{\ntop = population[as.scalar(selected[i]), ]\n@@ -140,7 +138,6 @@ return (Frame[Unknown] bestLg, Double pre_best, Double T)\n# # # append length of pipeline and pipeline in frame\n# #\n- print(\"problem kia he apka\")\nchildren[i, 1] = ncol(c1)\nchildren[i, 2:(ncol(c1) + 1)] = c1\ni = i + 1\n@@ -161,7 +158,6 @@ return (Frame[Unknown] bestLg, Double pre_best, Double T)\naddition = function(Frame[Unknown] top, Frame[Unknown] allOps, Integer addCount)\nreturn (Frame [Unknown] child)\n{\n- print(\"Starting addition\")\nfor(i in 1:addCount)\n{\nc = as.scalar(sample(ncol(allOps), 1))\n@@ -182,7 +178,6 @@ return (Frame [Unknown] child)\nmutation = function(Frame[Unknown] child, Double mutationRate)\nreturn (Frame [Unknown] mChild)\n{\n- print(\"Starting mutation on \"+toString(child))\nrandom = as.scalar(rand(rows=1, cols=1))\nif(random > mutationRate & ncol(child) >= 3)\n{\n@@ -201,7 +196,6 @@ return (Frame[Unknown] output)\n{\nif(ncol(child) > 2 & (ncol(child)-2) > removal & removal > 0)\n{\n- print(\"Starting removal on \"+toString(child))\nfor(i in 1:removal)\n{\nidx = as.scalar(sample(ncol(child)-3, 1))\n@@ -215,5 +209,4 @@ return (Frame[Unknown] output)\n}\n}\noutput = child\n- print(\"ended removal on \"+toString(output))\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkCleaningClassificationTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/pipelines/BuiltinTopkCleaningClassificationTest.java", "diff": "@@ -67,7 +67,6 @@ public class BuiltinTopkCleaningClassificationTest extends AutomatedTestBase {\nprivate void runtopkCleaning(Double sample, int topk, int resources, String cv, int cvk , double split, Types.ExecMode et) {\n- setOutputBuffering(true);\nTypes.ExecMode modeOld = setExecMode(et);\nString HOME = SCRIPT_DIR + TEST_DIR;\ntry {\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/intermediates/classification/bestAcc.csv", "new_path": "src/test/scripts/functions/pipelines/intermediates/classification/bestAcc.csv", "diff": "+85.58558558558559\n84.68468468468468\n82.88288288288288\n-82.88288288288288\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/intermediates/classification/hp.csv", "new_path": "src/test/scripts/functions/pipelines/intermediates/classification/hp.csv", "diff": "+36.0,3.0,3.0,2.0,1.0,0,0,0,1.0,0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,1.0,0,2.0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n36.0,3.0,2.0,2.0,1.0,0,0,0,1.0,0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,1.0,0,2.0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n-36.0,3.0,1.0,1.0,1.0,0,0,0,1.0,0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,1.0,0,2.0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n-36.0,3.0,7.0,2.0,1.0,0,0,0,1.0,0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,1.0,0,2.0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n+36.0,3.0,3.0,1.0,1.0,0,0,0,1.0,0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,1.0,0,2.0,0,0,0,0,1.0,0,0,0,2.0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/intermediates/classification/pip.csv", "new_path": "src/test/scripts/functions/pipelines/intermediates/classification/pip.csv", "diff": "outlierBySd,imputeByMedian,wtomeklink,dummycoding\n-outlierBySd,imputeByMean,wtomeklink,dummycoding\n+outlierBySd,imputeByMedian,wtomeklink,dummycoding\noutlierBySd,imputeByMedian,wtomeklink,dummycoding\n" }, { "change_type": "MODIFY", "old_path": "src/test/scripts/functions/pipelines/topkcleaningClassificationTest.dml", "new_path": "src/test/scripts/functions/pipelines/topkcleaningClassificationTest.dml", "diff": "@@ -70,10 +70,8 @@ evalClassification = function(Matrix[Double] X, Matrix[Double] Y, Matrix[Double]\nreturn(Matrix[Double] output)\n{\n- print(\"trainML: \"+as.integer(trainML))\nif(trainML == 1)\n{\n- print(\"training\")\nparams = list(\"icpt\", \"reg\", \"tol\", \"maxii\")\nparamRanges = list(seq(0, 2, 1), 10^seq(1,-3), 10^seq(1,-5), 10^seq(1,3));\ntrainArgs = list(X=X, Y=Y, icpt=-1, reg=-1, tol=-1, maxi=100, maxii=-1, verbose=FALSE);\n@@ -84,63 +82,13 @@ return(Matrix[Double] output)\nbeta = multiLogReg(X=X, Y=Y, icpt=as.scalar(evalFunHp[1,1]), reg=as.scalar(evalFunHp[1,2]), tol=as.scalar(evalFunHp[1,3]),\nmaxi=as.scalar(evalFunHp[1,4]), maxii=50, verbose=FALSE);\n[prob, yhat, accuracy] = multiLogRegPredict(Xtest, beta, Ytest, FALSE)\n- print(\"accuracy a: \"+toString(accuracy))\na = getAccuracy(Ytest, yhat, TRUE)\n- print(\"accuracy weighted: \"+a)\n+ print(\"accuracy: \"+toString(accuracy)+\" weighted accuracy: \"+a)\naccuracy = as.matrix(accuracy)\noutput = cbind(accuracy, evalFunHp)\n- print(\"output: \"+toString(output))\n-}\n-\n-# UDF for evaluation\n-# choice of parameters provided by API, X, Y, clone_X, evalFunHp (hyper-param), trainML (boolean for optimizing hp internally or passed by externally )\n-evalClassificationOLd = function(Matrix[Double] X, Matrix[Double] Y, Matrix[Double] Xtest, Matrix[Double] Ytest, Matrix[Double] Xorig, List[Unknown] pipList, List[Unknown] metaList,\n- Matrix[Double] evalFunHp, Integer trainML=0)\n-return(Matrix[Double] output)\n-{\n- score = as.double(0)\n- mask = as.matrix(metaList['mask'])\n- cv = FALSE\n- print(\"cols in X and Xtest: \"+ncol(X)+\" \"+ncol(Xtest))\n- if(ncol(X) != ncol(Xtest))\n- stop(\"Dimension mismatch: number of columns and train and test are not equal\")\n-\n- if(trainML == 1)\n- {\n- # do the gridsearch for hyper-parameters\n- params = list(\"icpt\", \"reg\", \"tol\", \"maxii\")\n- paramRanges = list(seq(0, 2, 1), 10^seq(1,-3), 10^seq(1,-5), 10^seq(1,3));\n- trainArgs = list(X=X, Y=Y, icpt=-1, reg=-1, tol=-1, maxi=100, maxii=-1, verbose=FALSE);\n- [B1, opt] = utils::topk_gridSearch(X=X, y=Y, Xtest=Xtest, ytest=Ytest, train=\"multiLogReg\", predict=\"accuracy\", numB=ncol(X)+1, cv=FALSE, cvk=0,\n- params=params, paramValues=paramRanges, trainArgs=trainArgs, verbose=FALSE);\n- evalFunHp = as.matrix(opt)\n- }\n- # do the hold out train/test\n- # evalFunHpM = as.matrix(evalFunHp)\n- if(as.scalar(pipList['flags']) != 0)\n- {\n- [X, Y, Xtest, Ytest, Tr] = executePipeline(as.frame(pipList['lp']), as.frame(pipList['ph']), X, Y, Xtest, Ytest, as.matrix(metaList['mask']), as.matrix(metaList['fd']),\n- as.matrix(pipList['hp']), as.scalar(pipList['flags']), TRUE, FALSE)\n- }\n- print(\"min and max of y in eval: \"+min(Y)+\" \"+max(Y))\n- if(max(Y) == min(Y)) {\n- print(\"Y contains only one class\")\n- }\n- else {\n- beta = multiLogReg(X=X, Y=Y, icpt=as.scalar(evalFunHp[1,1]), reg=as.scalar(evalFunHp[1,2]), tol=as.scalar(evalFunHp[1,3]),\n- maxi=as.scalar(evalFunHp[1,4]), maxii=50, verbose=FALSE);\n-\n- [prob, yhat, acc] = multiLogRegPredict(Xtest, beta, Ytest, FALSE)\n- score = getAccuracy(Ytest, yhat, TRUE)\n- }\n-\n- output = cbind(as.matrix(acc), evalFunHp)\n- print(\"hold out accuracy: \"+acc)\n- print(\"hold out waccuracy: \"+score)\n-\n}\naccuracy = function(Matrix[Double] X, Matrix[Double] y, Matrix[Double] B) return (Matrix[Double] err) {\n- [M,yhat,acc] = multiLogRegPredict(X=X, B=B, Y=y, verbose=TRUE);\n+ [M,yhat,acc] = multiLogRegPredict(X=X, B=B, Y=y, verbose=FALSE);\nerr = as.matrix(1-(acc/100));\n}\n\\ No newline at end of file\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/ipa/IPAPassCompressionWorkloadAnalysis.java", "new_path": "src/main/java/org/apache/sysds/hops/ipa/IPAPassCompressionWorkloadAnalysis.java", "diff": "@@ -66,6 +66,5 @@ public class IPAPassCompressionWorkloadAnalysis extends IPAPass {\n}\nreturn map != null;\n-\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/rewrite/RewriteCompressedReblock.java", "new_path": "src/main/java/org/apache/sysds/hops/rewrite/RewriteCompressedReblock.java", "diff": "@@ -141,21 +141,17 @@ public class RewriteCompressedReblock extends StatementBlockRewriteRule {\nreturn satisfies;\n}\n- private static boolean satisfiesAggressiveCompressionCondition(Hop hop) {\n- boolean satisfies = false;\n+ public static boolean satisfiesAggressiveCompressionCondition(Hop hop) {\n+ //size-independent conditions (robust against unknowns)\n+ boolean satisfies = HopRewriteUtils.isTernary(hop, OpOp3.CTABLE) //matrix (no vector) ctable\n+ && hop.getInput(0).getDataType().isMatrix() && hop.getInput(1).getDataType().isMatrix();\n+ //size-dependent conditions\nif(satisfiesSizeConstraintsForCompression(hop)) {\nsatisfies |= HopRewriteUtils.isData(hop, OpOpData.PERSISTENTREAD);\nsatisfies |= HopRewriteUtils.isUnary(hop, OpOp1.ROUND, OpOp1.FLOOR, OpOp1.NOT, OpOp1.CEIL);\n- satisfies |= HopRewriteUtils.isBinary(hop,\n- OpOp2.EQUAL,\n- OpOp2.NOTEQUAL,\n- OpOp2.LESS,\n- OpOp2.LESSEQUAL,\n- OpOp2.GREATER,\n- OpOp2.GREATEREQUAL,\n- OpOp2.AND,\n- OpOp2.OR,\n- OpOp2.MODULUS);\n+ satisfies |= HopRewriteUtils.isBinary(hop, OpOp2.EQUAL, OpOp2.NOTEQUAL, OpOp2.LESS,\n+ OpOp2.LESSEQUAL, OpOp2.GREATER, OpOp2.GREATEREQUAL, OpOp2.AND, OpOp2.OR, OpOp2.MODULUS);\n+ satisfies |= HopRewriteUtils.isTernary(hop, OpOp3.CTABLE);\n}\nif(LOG.isDebugEnabled() && satisfies)\nLOG.debug(\"Operation Satisfies: \" + hop);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/dictionary/MatrixBlockDictionary.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/dictionary/MatrixBlockDictionary.java", "diff": "@@ -24,7 +24,6 @@ import java.io.DataOutput;\nimport java.io.IOException;\nimport org.apache.commons.lang.NotImplementedException;\n-import org.apache.sysds.runtime.compress.DMLCompressionException;\nimport org.apache.sysds.runtime.data.DenseBlock;\nimport org.apache.sysds.runtime.data.DenseBlockFP64;\nimport org.apache.sysds.runtime.data.SparseBlock;\n@@ -50,7 +49,11 @@ public class MatrixBlockDictionary extends ADictionary {\n@Override\npublic double[] getValues() {\n- throw new DMLCompressionException(\"Get Values should not be called when you have a MatrixBlockDictionary\");\n+ // FIXME fix MinMaxGroup Initialization to avoid conversion to dense\n+ if( !_data.isInSparseFormat() )\n+ _data.sparseToDense();\n+ return _data.getDenseBlockValues();\n+ //throw new DMLCompressionException(\"Get Values should not be called when you have a MatrixBlockDictionary\");\n}\n@Override\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/workload/WorkloadAnalyzer.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/workload/WorkloadAnalyzer.java", "diff": "@@ -23,6 +23,7 @@ import java.util.ArrayList;\nimport java.util.HashMap;\nimport java.util.HashSet;\nimport java.util.Iterator;\n+import java.util.LinkedList;\nimport java.util.List;\nimport java.util.Map;\nimport java.util.Set;\n@@ -58,6 +59,11 @@ import org.apache.sysds.runtime.compress.workload.AWTreeNode.WTNodeType;\npublic class WorkloadAnalyzer {\nprivate static final Log LOG = LogFactory.getLog(WorkloadAnalyzer.class.getName());\n+ // indicator for more aggressive compression of intermediates\n+ public static boolean ALLOW_INTERMEDIATE_CANDIDATES = false;\n+ // avoid wtree construction for assumptionly already compressed intermediates\n+ // (due to conditional control flow this might miss compression opportunities)\n+ public static boolean PRUNE_COMPRESSED_INTERMEDIATES = true;\nprivate final Set<Hop> visited;\nprivate final Set<Long> compressed;\n@@ -69,16 +75,23 @@ public class WorkloadAnalyzer {\nprivate final List<Hop> decompressHops;\npublic static Map<Long, WTreeRoot> getAllCandidateWorkloads(DMLProgram prog) {\n- // extract all compression candidates from program\n+ // extract all compression candidates from program (in program order)\nList<Hop> candidates = getCandidates(prog);\n// for each candidate, create pruned workload tree\n- // TODO memoization of processed subtree if overlap\n+ List<WorkloadAnalyzer> allWAs = new LinkedList<>();\nMap<Long, WTreeRoot> map = new HashMap<>();\nfor(Hop cand : candidates) {\n- WTreeRoot tree = new WorkloadAnalyzer(prog).createWorkloadTree(cand);\n-\n+ //prune already covered candidate (intermediate already compressed)\n+ if( PRUNE_COMPRESSED_INTERMEDIATES )\n+ if( allWAs.stream().anyMatch(w -> w.containsCompressed(cand)) )\n+ continue; //intermediate already compressed\n+\n+ //construct workload tree for candidate\n+ WorkloadAnalyzer wa = new WorkloadAnalyzer(prog);\n+ WTreeRoot tree = wa.createWorkloadTree(cand);\nmap.put(cand.getHopID(), tree);\n+ allWAs.add(wa);\n}\nreturn map;\n@@ -128,6 +141,10 @@ public class WorkloadAnalyzer {\nreturn main;\n}\n+ protected boolean containsCompressed(Hop hop) {\n+ return compressed.contains(hop.getHopID());\n+ }\n+\nprivate static List<Hop> getCandidates(DMLProgram prog) {\nList<Hop> candidates = new ArrayList<>();\nfor(StatementBlock sb : prog.getStatementBlocks()) {\n@@ -191,7 +208,9 @@ public class WorkloadAnalyzer {\nif(hop.isVisited())\nreturn;\n// evaluate and add candidates (type and size)\n- if(RewriteCompressedReblock.satisfiesCompressionCondition(hop))\n+ if( ( RewriteCompressedReblock.satisfiesAggressiveCompressionCondition(hop)\n+ & ALLOW_INTERMEDIATE_CANDIDATES)\n+ || RewriteCompressedReblock.satisfiesCompressionCondition(hop))\ncands.add(hop);\n// recursively process children (inputs)\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/compress/workload/WorkloadAlgorithmTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/compress/workload/WorkloadAlgorithmTest.java", "diff": "@@ -24,6 +24,7 @@ import static org.junit.Assert.fail;\nimport java.io.File;\nimport org.apache.sysds.common.Types.ExecMode;\n+import org.apache.sysds.runtime.compress.workload.WorkloadAnalyzer;\nimport org.apache.sysds.test.AutomatedTestBase;\nimport org.apache.sysds.test.TestConfiguration;\nimport org.apache.sysds.test.TestUtils;\n@@ -36,6 +37,7 @@ public class WorkloadAlgorithmTest extends AutomatedTestBase {\nprivate final static String TEST_NAME1 = \"WorkloadAnalysisMLogReg\";\nprivate final static String TEST_NAME2 = \"WorkloadAnalysisLm\";\nprivate final static String TEST_NAME3 = \"WorkloadAnalysisPCA\";\n+ private final static String TEST_NAME4 = \"WorkloadAnalysisSliceLine\";\nprivate final static String TEST_DIR = \"functions/compress/workload/\";\nprivate final static String TEST_CLASS_DIR = TEST_DIR + WorkloadAnalysisTest.class.getSimpleName() + \"/\";\n@@ -45,39 +47,51 @@ public class WorkloadAlgorithmTest extends AutomatedTestBase {\naddTestConfiguration(TEST_NAME1, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME1, new String[] {\"B\"}));\naddTestConfiguration(TEST_NAME2, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME2, new String[] {\"B\"}));\naddTestConfiguration(TEST_NAME3, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME3, new String[] {\"B\"}));\n-\n+ addTestConfiguration(TEST_NAME4, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME4, new String[] {\"B\"}));\n}\n@Test\npublic void testMLogRegCP() {\n- runWorkloadAnalysisTest(TEST_NAME1, ExecMode.HYBRID, 2);\n+ runWorkloadAnalysisTest(TEST_NAME1, ExecMode.HYBRID, 2, false);\n}\n@Test\npublic void testLmSP() {\n- runWorkloadAnalysisTest(TEST_NAME2, ExecMode.SPARK, 2);\n+ runWorkloadAnalysisTest(TEST_NAME2, ExecMode.SPARK, 2, false);\n}\n@Test\npublic void testLmCP() {\n- runWorkloadAnalysisTest(TEST_NAME2, ExecMode.HYBRID, 2);\n+ runWorkloadAnalysisTest(TEST_NAME2, ExecMode.HYBRID, 2, false);\n}\n@Test\npublic void testPCASP() {\n- runWorkloadAnalysisTest(TEST_NAME3, ExecMode.SPARK, 1);\n+ runWorkloadAnalysisTest(TEST_NAME3, ExecMode.SPARK, 1, false);\n}\n@Test\npublic void testPCACP() {\n- runWorkloadAnalysisTest(TEST_NAME3, ExecMode.HYBRID, 1);\n+ runWorkloadAnalysisTest(TEST_NAME3, ExecMode.HYBRID, 1, false);\n+ }\n+\n+ @Test\n+ public void testSliceLineCP1() {\n+ runWorkloadAnalysisTest(TEST_NAME4, ExecMode.HYBRID, 0, false);\n}\n- private void runWorkloadAnalysisTest(String testname, ExecMode mode, int compressionCount) {\n+ @Test\n+ public void testSliceLineCP2() {\n+ runWorkloadAnalysisTest(TEST_NAME4, ExecMode.HYBRID, 2, true);\n+ }\n+\n+ private void runWorkloadAnalysisTest(String testname, ExecMode mode, int compressionCount, boolean intermediates) {\nExecMode oldPlatform = setExecMode(mode);\n+ boolean oldIntermediates = WorkloadAnalyzer.ALLOW_INTERMEDIATE_CANDIDATES;\ntry {\nloadTestConfiguration(getTestConfiguration(testname));\n+ WorkloadAnalyzer.ALLOW_INTERMEDIATE_CANDIDATES = intermediates;\nString HOME = SCRIPT_DIR + TEST_DIR;\nfullDMLScriptName = HOME + testname + \".dml\";\n@@ -99,9 +113,11 @@ public class WorkloadAlgorithmTest extends AutomatedTestBase {\n.getCPHeavyHitterCount(\"compress\") : Statistics.getCPHeavyHitterCount(\"sp_compress\");\nAssert.assertEquals(compressionCount, actualCompressionCount);\n- Assert.assertTrue( mode == ExecMode.HYBRID ? heavyHittersContainsString(\"compress\") : heavyHittersContainsString(\"sp_compress\"));\n+ if( compressionCount > 0 )\n+ Assert.assertTrue( mode == ExecMode.HYBRID ?\n+ heavyHittersContainsString(\"compress\") : heavyHittersContainsString(\"sp_compress\"));\n+ if( !testname.equals(TEST_NAME4) )\nAssert.assertFalse(heavyHittersContainsString(\"m_scale\"));\n-\n}\ncatch(Exception e) {\nresetExecMode(oldPlatform);\n@@ -109,6 +125,7 @@ public class WorkloadAlgorithmTest extends AutomatedTestBase {\n}\nfinally {\nresetExecMode(oldPlatform);\n+ WorkloadAnalyzer.ALLOW_INTERMEDIATE_CANDIDATES = oldIntermediates;\n}\n}\n" }, { "change_type": "RENAME", "old_path": "src/test/scripts/functions/compress/workload/WorkloadAnalysisSliceFinder.dml", "new_path": "src/test/scripts/functions/compress/workload/WorkloadAnalysisSliceLine.dml", "diff": "#\n#-------------------------------------------------------------\n-X = read($1) + 1;\n-Y = read($2);\n-\n-\n-print(\"\")\n-print(\"MLogReg\")\n-\n-[X_s,s,c] = scale(X=X, scale=TRUE, center=TRUE);\n-B = multiLogReg(X=X_s, Y=Y, verbose=FALSE, maxi=2, maxii=2);\n-[nn, P, acc] = multiLogRegPredict(X=X_s, B=B, Y=Y)\n-\n-\n-[nn, C] = confusionMatrix(P, Y)\n-print(\"Confusion: \")\n-print(toString(C))\n-\n-\n-print(\"\")\n-print(\"SliceFinder\")\n-\n-e = Y == P\n-\n-[tk,tkc,d] = slicefinder(X=X, e=e, maxL = 2, verbose=TRUE)\n-\n-print(\"tk :\\n\" + toString(tk))\n-print(\"tkc :\\n\" + toString(tkc))\n+# data preparation\n+FXY = read(\"./src/test/resources/datasets/Salaries.csv\",\n+ data_type=\"frame\", format=\"csv\", header=TRUE);\n+F = FXY[,1:ncol(FXY)-1];\n+y = as.matrix(FXY[,ncol(FXY)]);\n+jspec= \"{ ids:true, recode:[1,2,3,6],bin:[\"\n+ +\"{id:4, method:equi-width, numbins:14},\"\n+ +\"{id:5, method:equi-width, numbins:12}]}\"\n+[X,M] = transformencode(target=F, spec=jspec);\n+X = X[,2:ncol(X)]\n+\n+m = nrow(X)\n+n = ncol(X)\n+fdom = colMaxs(X);\n+foffb = t(cumsum(t(fdom))) - fdom;\n+foffe = t(cumsum(t(fdom)))\n+rix = matrix(seq(1,m)%*%matrix(1,1,n), m*n, 1)\n+cix = matrix(X + foffb, m*n, 1);\n+X2 = table(rix, cix); #one-hot encoded\n+\n+# learn model\n+B = lm(X=X2, y=y, verbose=FALSE);\n+yhat = X2 %*% B;\n+e = (y-yhat)^2;\n+\n+# call slice finding\n+[TS,TR,d] = slicefinder(X=X, e=e, k=10,\n+ alpha=0.95, minSup=4, tpEval=TRUE, verbose=TRUE);\n+\n+print(\"TS:\\n\" + toString(TS))\n+print(\"TR:\\n\" + toString(TR))\nprint(\"Debug matrix:\\n\" + toString(d))\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/recompile/Recompiler.java", "new_path": "src/main/java/org/apache/sysds/hops/recompile/Recompiler.java", "diff": "@@ -93,6 +93,7 @@ import org.apache.sysds.runtime.instructions.cp.IntObject;\nimport org.apache.sysds.runtime.instructions.cp.ListObject;\nimport org.apache.sysds.runtime.instructions.cp.ScalarObject;\nimport org.apache.sysds.runtime.io.IOUtilFunctions;\n+import org.apache.sysds.runtime.lineage.LineageItem;\nimport org.apache.sysds.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysds.runtime.meta.DataCharacteristics;\nimport org.apache.sysds.runtime.meta.MatrixCharacteristics;\n@@ -1593,7 +1594,8 @@ public class Recompiler {\nlong estFilesize = (long)(3.5 * mem); //conservative estimate\nlong cpThreshold = CP_REBLOCK_THRESHOLD_SIZE *\nOptimizerUtils.getParallelTextReadParallelism();\n- return (estFilesize < cpThreshold);\n+ return (iimd.getFileFormat() == FileFormat.BINARY\n+ || estFilesize < cpThreshold); //for text conservative\n}\npublic static boolean checkCPCheckpoint(DataCharacteristics dc) {\n@@ -1602,8 +1604,12 @@ public class Recompiler {\n&& !OptimizerUtils.exceedsCachingThreshold(dc.getCols(), OptimizerUtils.estimateSize(dc));\n}\n- @SuppressWarnings(\"unchecked\")\npublic static void executeInMemoryReblock(ExecutionContext ec, String varin, String varout) {\n+ executeInMemoryReblock(ec, varin, varout, null);\n+ }\n+\n+ @SuppressWarnings(\"unchecked\")\n+ public static void executeInMemoryReblock(ExecutionContext ec, String varin, String varout, LineageItem litem) {\nCacheableData<CacheBlock> in = (CacheableData<CacheBlock>) ec.getCacheableData(varin);\nCacheableData<CacheBlock> out = (CacheableData<CacheBlock>) ec.getCacheableData(varout);\n@@ -1618,6 +1624,7 @@ public class Recompiler {\n//set output (incl update matrix characteristics)\nout.acquireModify(mb);\n+ out.setCacheLineage(litem);\nout.release();\nin.release();\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/CSVReblockSPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/CSVReblockSPInstruction.java", "diff": "@@ -119,8 +119,9 @@ public class CSVReblockSPInstruction extends UnarySPInstruction {\n//check for in-memory reblock (w/ lazy spark context, potential for latency reduction)\nif( Recompiler.checkCPReblock(sec, input1.getName()) ) {\n- if( input1.getDataType().isMatrix() || input1.getDataType().isFrame() )\n+ if( input1.getDataType().isMatrix() || input1.getDataType().isFrame() ) {\nRecompiler.executeInMemoryReblock(sec, input1.getName(), output.getName());\n+ }\nStatistics.decrementNoOfExecutedSPInst();\nreturn;\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/ReblockSPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/spark/ReblockSPInstruction.java", "diff": "@@ -21,6 +21,7 @@ package org.apache.sysds.runtime.instructions.spark;\nimport java.util.Set;\n+import org.apache.commons.lang3.tuple.Pair;\nimport org.apache.hadoop.io.LongWritable;\nimport org.apache.hadoop.io.Text;\nimport org.apache.spark.api.java.JavaPairRDD;\n@@ -41,6 +42,7 @@ import org.apache.sysds.runtime.io.FileFormatPropertiesCSV;\nimport org.apache.sysds.runtime.io.FileFormatPropertiesLIBSVM;\nimport org.apache.sysds.runtime.io.FileFormatPropertiesMM;\nimport org.apache.sysds.runtime.io.IOUtilFunctions;\n+import org.apache.sysds.runtime.lineage.LineageItem;\nimport org.apache.sysds.runtime.matrix.data.FrameBlock;\nimport org.apache.sysds.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysds.runtime.matrix.data.MatrixCell;\n@@ -48,6 +50,7 @@ import org.apache.sysds.runtime.matrix.data.MatrixIndexes;\nimport org.apache.sysds.runtime.matrix.operators.Operator;\nimport org.apache.sysds.runtime.meta.DataCharacteristics;\nimport org.apache.sysds.runtime.meta.MetaDataFormat;\n+import org.apache.sysds.runtime.util.ProgramConverter;\nimport org.apache.sysds.utils.Statistics;\npublic class ReblockSPInstruction extends UnarySPInstruction {\n@@ -96,8 +99,10 @@ public class ReblockSPInstruction extends UnarySPInstruction {\n//check for in-memory reblock (w/ lazy spark context, potential for latency reduction)\nif( Recompiler.checkCPReblock(sec, input1.getName()) ) {\n- if( input1.getDataType().isMatrix() || input1.getDataType().isFrame() )\n- Recompiler.executeInMemoryReblock(sec, input1.getName(), output.getName());\n+ if( input1.getDataType().isMatrix() || input1.getDataType().isFrame() ) {\n+ Recompiler.executeInMemoryReblock(sec, input1.getName(), output.getName(),\n+ iimd.getFileFormat()==FileFormat.BINARY ? getLineageItem(ec).getValue() : null);\n+ }\nStatistics.decrementNoOfExecutedSPInst();\nreturn;\n}\n@@ -256,4 +261,15 @@ public class ReblockSPInstruction extends UnarySPInstruction {\n+ \"for ReblockSPInstruction: \" + fmt.toString());\n}\n}\n+\n+ @Override\n+ public Pair<String, LineageItem> getLineageItem(ExecutionContext ec) {\n+ //construct reblock lineage without existing createvar lineage\n+ if( ec.getLineage() == null ) {\n+ return Pair.of(output.getName(), new LineageItem(\n+ ProgramConverter.serializeDataObject(input1.getName(), ec.getCacheableData(input1)), \"cache_rblk\"));\n+ }\n+ //default reblock w/ active lineage tracing\n+ return super.getLineageItem(ec);\n+ }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/lineage/LineageRecomputeUtils.java", "new_path": "src/main/java/org/apache/sysds/runtime/lineage/LineageRecomputeUtils.java", "diff": "@@ -31,6 +31,7 @@ import java.util.stream.Collectors;\nimport org.apache.commons.lang3.mutable.MutableInt;\nimport org.apache.sysds.api.DMLScript;\nimport org.apache.sysds.common.Types.DataType;\n+import org.apache.sysds.common.Types.FileFormat;\nimport org.apache.sysds.common.Types.OpOp1;\nimport org.apache.sysds.common.Types.OpOp2;\nimport org.apache.sysds.common.Types.OpOp3;\n@@ -58,6 +59,7 @@ import org.apache.sysds.runtime.DMLRuntimeException;\nimport org.apache.sysds.runtime.controlprogram.BasicProgramBlock;\nimport org.apache.sysds.runtime.controlprogram.FunctionProgramBlock;\nimport org.apache.sysds.runtime.controlprogram.Program;\n+import org.apache.sysds.runtime.controlprogram.caching.CacheableData;\nimport org.apache.sysds.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysds.runtime.controlprogram.context.ExecutionContextFactory;\nimport org.apache.sysds.runtime.instructions.Instruction;\n@@ -71,6 +73,7 @@ import org.apache.sysds.runtime.instructions.cp.ScalarObjectFactory;\nimport org.apache.sysds.runtime.instructions.cp.VariableCPInstruction;\nimport org.apache.sysds.runtime.instructions.spark.RandSPInstruction;\nimport org.apache.sysds.runtime.instructions.spark.SPInstruction.SPType;\n+import org.apache.sysds.runtime.util.ProgramConverter;\nimport org.apache.sysds.utils.Explain;\nimport org.apache.sysds.utils.Explain.ExplainCounts;\nimport org.apache.sysds.utils.Statistics;\n@@ -200,6 +203,20 @@ public class LineageRecomputeUtils {\noperands.put(item.getId(), input); // order preserving\nbreak;\n}\n+ else if( item.getOpcode().equals(\"cache_rblk\") ) {\n+ CacheableData<?> dat = (CacheableData<?>)ProgramConverter.parseDataObject(item.getData())[1];\n+ DataOp hop = new DataOp(\"tmp\", dat.getDataType(), dat.getValueType(),\n+ OpOpData.PERSISTENTREAD, dat.getFileName(), dat.getNumRows(),\n+ dat.getNumColumns(), dat.getDataCharacteristics().getNonZeros(), -1);\n+ hop.setFileFormat(FileFormat.BINARY);\n+ hop.setInputBlocksize(dat.getBlocksize());\n+ hop.setBlocksize(ConfigurationManager.getBlocksize());\n+ hop.setRequiresReblock(true);\n+ operands.put(item.getId(), hop);\n+ break;\n+ }\n+\n+\nInstruction inst = InstructionParser.parseSingleInstruction(item.getData());\nif (inst instanceof DataGenCPInstruction) {\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/data/SparseRowScalar.java", "new_path": "src/main/java/org/apache/sysds/runtime/data/SparseRowScalar.java", "diff": "package org.apache.sysds.runtime.data;\n-import java.io.Serializable;\n-\n-public final class SparseRowScalar extends SparseRow implements Serializable\n-{\n+public final class SparseRowScalar extends SparseRow{\nprivate static final long serialVersionUID = 722193514969067477L;\nprivate int index;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/data/SparseRowVector.java", "new_path": "src/main/java/org/apache/sysds/runtime/data/SparseRowVector.java", "diff": "package org.apache.sysds.runtime.data;\n-import java.io.Serializable;\nimport java.util.Arrays;\nimport org.apache.sysds.runtime.util.SortUtils;\nimport org.apache.sysds.runtime.util.UtilFunctions;\n-public final class SparseRowVector extends SparseRow implements Serializable\n-{\n+public final class SparseRowVector extends SparseRow{\nprivate static final long serialVersionUID = 2971077474424464992L;\n//initial capacity of any created sparse row\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/matrix/operators/AggregateOperator.java", "new_path": "src/main/java/org/apache/sysds/runtime/matrix/operators/AggregateOperator.java", "diff": "package org.apache.sysds.runtime.matrix.operators;\n-import java.io.Serializable;\n-\nimport org.apache.sysds.common.Types.CorrectionLocationType;\nimport org.apache.sysds.runtime.functionobjects.KahanPlus;\nimport org.apache.sysds.runtime.functionobjects.KahanPlusSq;\n@@ -31,8 +29,7 @@ import org.apache.sysds.runtime.functionobjects.Plus;\nimport org.apache.sysds.runtime.functionobjects.ValueFunction;\n-public class AggregateOperator extends Operator implements Serializable\n-{\n+public class AggregateOperator extends Operator {\nprivate static final long serialVersionUID = 8761527329665129670L;\npublic final double initialValue;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/matrix/operators/AggregateTernaryOperator.java", "new_path": "src/main/java/org/apache/sysds/runtime/matrix/operators/AggregateTernaryOperator.java", "diff": "package org.apache.sysds.runtime.matrix.operators;\n-import java.io.Serializable;\n-\nimport org.apache.sysds.runtime.functionobjects.IndexFunction;\nimport org.apache.sysds.runtime.functionobjects.ValueFunction;\n-public class AggregateTernaryOperator extends Operator implements Serializable\n+public class AggregateTernaryOperator extends Operator\n{\nprivate static final long serialVersionUID = 4251745081160216784L;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/matrix/operators/AggregateUnaryOperator.java", "new_path": "src/main/java/org/apache/sysds/runtime/matrix/operators/AggregateUnaryOperator.java", "diff": "@@ -30,8 +30,7 @@ import org.apache.sysds.runtime.functionobjects.ReduceCol;\nimport org.apache.sysds.runtime.functionobjects.ReduceRow;\n-public class AggregateUnaryOperator extends Operator\n-{\n+public class AggregateUnaryOperator extends Operator {\nprivate static final long serialVersionUID = 6690553323120787735L;\npublic final AggregateOperator aggOp;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/matrix/operators/BinaryOperator.java", "new_path": "src/main/java/org/apache/sysds/runtime/matrix/operators/BinaryOperator.java", "diff": "package org.apache.sysds.runtime.matrix.operators;\n-import java.io.Serializable;\n-\nimport org.apache.sysds.common.Types.OpOp2;\nimport org.apache.sysds.runtime.functionobjects.And;\nimport org.apache.sysds.runtime.functionobjects.BitwAnd;\n@@ -30,6 +28,7 @@ import org.apache.sysds.runtime.functionobjects.BitwShiftL;\nimport org.apache.sysds.runtime.functionobjects.BitwShiftR;\nimport org.apache.sysds.runtime.functionobjects.BitwXor;\nimport org.apache.sysds.runtime.functionobjects.Builtin;\n+import org.apache.sysds.runtime.functionobjects.Builtin.BuiltinCode;\nimport org.apache.sysds.runtime.functionobjects.Divide;\nimport org.apache.sysds.runtime.functionobjects.Equals;\nimport org.apache.sysds.runtime.functionobjects.GreaterThan;\n@@ -50,10 +49,8 @@ import org.apache.sysds.runtime.functionobjects.PlusMultiply;\nimport org.apache.sysds.runtime.functionobjects.Power;\nimport org.apache.sysds.runtime.functionobjects.ValueFunction;\nimport org.apache.sysds.runtime.functionobjects.Xor;\n-import org.apache.sysds.runtime.functionobjects.Builtin.BuiltinCode;\n-public class BinaryOperator extends Operator implements Serializable\n-{\n+public class BinaryOperator extends Operator {\nprivate static final long serialVersionUID = -2547950181558989209L;\npublic final ValueFunction fn;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/matrix/operators/ReorgOperator.java", "new_path": "src/main/java/org/apache/sysds/runtime/matrix/operators/ReorgOperator.java", "diff": "package org.apache.sysds.runtime.matrix.operators;\n-import java.io.Serializable;\n-\nimport org.apache.sysds.runtime.functionobjects.IndexFunction;\n-public class ReorgOperator extends Operator implements Serializable\n-{\n+public class ReorgOperator extends Operator{\nprivate static final long serialVersionUID = -5322516429026298404L;\npublic final IndexFunction fn;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/matrix/operators/TernaryOperator.java", "new_path": "src/main/java/org/apache/sysds/runtime/matrix/operators/TernaryOperator.java", "diff": "package org.apache.sysds.runtime.matrix.operators;\n-import java.io.Serializable;\n-\nimport org.apache.sysds.runtime.functionobjects.IfElse;\nimport org.apache.sysds.runtime.functionobjects.MinusMultiply;\nimport org.apache.sysds.runtime.functionobjects.PlusMultiply;\nimport org.apache.sysds.runtime.functionobjects.TernaryValueFunction;\n-public class TernaryOperator extends Operator implements Serializable\n-{\n+public class TernaryOperator extends Operator{\nprivate static final long serialVersionUID = 3456088891054083634L;\npublic final TernaryValueFunction fn;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/data/DenseBlockLDRB.java", "new_path": "src/main/java/org/apache/sysds/runtime/data/DenseBlockLDRB.java", "diff": "@@ -72,7 +72,7 @@ public abstract class DenseBlockLDRB extends DenseBlock\nint lastBlockSize = (newBlockSize == rlen ? newBlockSize : rlen % newBlockSize) * odims[0];\nallocateBlocks(numBlocks);\nIntStream.range(0, numBlocks)\n- .forEach((i) -> {\n+ .forEach(i -> {\nint length = (i == numBlocks - 1 ? lastBlockSize : newBlockSize * _odims[0]);\nallocateBlock(i, length);\nif (v != 0)\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/io/ReaderTextCSVParallel.java", "new_path": "src/main/java/org/apache/sysds/runtime/io/ReaderTextCSVParallel.java", "diff": "@@ -40,6 +40,7 @@ import org.apache.hadoop.mapred.TextInputFormat;\nimport org.apache.sysds.conf.ConfigurationManager;\nimport org.apache.sysds.hops.OptimizerUtils;\nimport org.apache.sysds.runtime.DMLRuntimeException;\n+import org.apache.sysds.runtime.data.DenseBlock;\nimport org.apache.sysds.runtime.data.SparseBlock;\nimport org.apache.sysds.runtime.data.SparseRow;\nimport org.apache.sysds.runtime.io.IOUtilFunctions.CountRowsTask;\n@@ -93,7 +94,7 @@ public class ReaderTextCSVParallel extends MatrixReader {\n// allocate output matrix block\n// First Read Pass (count rows/cols, determine offsets, allocate matrix block)\n- MatrixBlock ret = computeCSVSizeAndCreateOutputMatrixBlock(splits, path, rlen, clen, estnnz);\n+ MatrixBlock ret = computeCSVSizeAndCreateOutputMatrixBlock(splits, path, rlen, clen, blen, estnnz);\n// Second Read Pass (read, parse strings, append to matrix block)\nreadCSVMatrixFromHDFS(splits, path, ret);\n@@ -156,8 +157,8 @@ public class ReaderTextCSVParallel extends MatrixReader {\n}\n}\n- private MatrixBlock computeCSVSizeAndCreateOutputMatrixBlock(InputSplit[] splits, Path path, long rlen, long clen,\n- long estnnz) throws IOException, DMLRuntimeException {\n+ private MatrixBlock computeCSVSizeAndCreateOutputMatrixBlock(InputSplit[] splits,\n+ Path path, long rlen, long clen, int blen, long estnnz) throws IOException, DMLRuntimeException {\n_rLen = 0;\n_cLen = 0;\n@@ -225,7 +226,7 @@ public class ReaderTextCSVParallel extends MatrixReader {\n// allocate target matrix block based on given size;\n// need to allocate sparse as well since lock-free insert into target\nlong estnnz2 = (estnnz < 0) ? (long) _rLen * _cLen : estnnz;\n- return createOutputMatrixBlock(_rLen, _cLen, _rLen, estnnz2, true, true);\n+ return createOutputMatrixBlock(_rLen, _cLen, blen, estnnz2, true, true);\n}\nprivate static class SplitOffsetInfos {\n@@ -262,8 +263,8 @@ public class ReaderTextCSVParallel extends MatrixReader {\nprotected final boolean _isFirstSplit;\nprotected final int _splitCount;\n- protected int row = 0;\n- protected int col = 0;\n+ protected int _row = 0;\n+ protected int _col = 0;\npublic CSVReadTask(InputSplit split, TextInputFormat informat, MatrixBlock dest, int splitCount) {\n_split = split;\n@@ -286,7 +287,7 @@ public class ReaderTextCSVParallel extends MatrixReader {\nreader.next(key, value);\n}\n- row = _offsets.getOffsetPerSplit(_splitCount);\n+ _row = _offsets.getOffsetPerSplit(_splitCount);\nlong nnz = 0;\ntry {\n@@ -300,8 +301,8 @@ public class ReaderTextCSVParallel extends MatrixReader {\n}\ncatch(Exception ex) {\n// post-mortem error handling and bounds checking\n- if(row < 0 || row + 1 > _rLen || col < 0 || col + 1 > _cLen) {\n- String errMsg = \"CSV cell [\" + (row + 1) + \",\" + (col + 1) + \"] \"\n+ if(_row < 0 || _row + 1 > _rLen || _col < 0 || _col + 1 > _cLen) {\n+ String errMsg = \"CSV cell [\" + (_row + 1) + \",\" + (_col + 1) + \"] \"\n+ \"out of overall matrix range [1:\" + _rLen + \",1:\" + _cLen + \"]. \" + ex.getMessage();\nthrow new IOException(errMsg, ex);\n}\n@@ -317,8 +318,8 @@ public class ReaderTextCSVParallel extends MatrixReader {\nthrows IOException;\nprotected void verifyRows(Text value) throws IOException {\n- if(row != (_offsets.getOffsetPerSplit(_splitCount) + _offsets.getLenghtPerSplit(_splitCount))) {\n- throw new IOException(\"Incorrect number of rows (\" + row + \") found in delimited file (\"\n+ if(_row != (_offsets.getOffsetPerSplit(_splitCount) + _offsets.getLenghtPerSplit(_splitCount))) {\n+ throw new IOException(\"Incorrect number of rows (\" + _row + \") found in delimited file (\"\n+ (_offsets.getOffsetPerSplit(_splitCount) + _offsets.getLenghtPerSplit(_splitCount)) + \"): \"\n+ value);\n}\n@@ -332,17 +333,18 @@ public class ReaderTextCSVParallel extends MatrixReader {\n}\nprotected long parse(RecordReader<LongWritable, Text> reader, LongWritable key, Text value) throws IOException {\n- double[] a = _dest.getDenseBlockValues();\n+ DenseBlock a = _dest.getDenseBlock();\ndouble cellValue = 0;\nlong nnz = 0;\nboolean noFillEmpty = false;\n- int index = row * (int) _cLen;\nwhile(reader.next(key, value)) { // foreach line\nfinal String cellStr = value.toString().trim();\nfinal String[] parts = IOUtilFunctions.split(cellStr, _props.getDelim());\n- for(String part : parts) { // foreach cell\n- part = part.trim();\n+ double[] avals = a.values(_row);\n+ int apos = a.pos(_row);\n+ for(int j = 0; j < _cLen; j++) { // foreach cell\n+ String part = parts[j].trim();\nif(part.isEmpty()) {\nnoFillEmpty |= !_props.isFill();\ncellValue = _props.getFillValue();\n@@ -351,15 +353,14 @@ public class ReaderTextCSVParallel extends MatrixReader {\ncellValue = Double.parseDouble(part);\n}\nif(cellValue != 0) {\n- a[index] = cellValue;\n+ avals[apos+j] = cellValue;\nnnz++;\n}\n- index++;\n}\n// sanity checks (number of columns, fill values)\nIOUtilFunctions.checkAndRaiseErrorCSVEmptyField(cellStr, _props.isFill(), noFillEmpty);\nIOUtilFunctions.checkAndRaiseErrorCSVNumColumns(_split.toString(), cellStr, parts, _cLen);\n- row++;\n+ _row++;\n}\nreturn nnz;\n@@ -374,16 +375,17 @@ public class ReaderTextCSVParallel extends MatrixReader {\n}\nprotected long parse(RecordReader<LongWritable, Text> reader, LongWritable key, Text value) throws IOException {\n- double[] a = _dest.getDenseBlockValues();\n+ DenseBlock a = _dest.getDenseBlock();\ndouble cellValue = 0;\nboolean noFillEmpty = false;\n- int index = row * (int) _cLen;\nlong nnz = 0;\nwhile(reader.next(key, value)) { // foreach line\nString cellStr = value.toString().trim();\nString[] parts = IOUtilFunctions.split(cellStr, _props.getDelim());\n- for(String part : parts) { // foreach cell\n- part = part.trim();\n+ double[] avals = a.values(_row);\n+ int apos = a.pos(_row);\n+ for(int j = 0; j < _cLen; j++) { // foreach cell\n+ String part = parts[j].trim();\nif(part.isEmpty()) {\nnoFillEmpty |= !_props.isFill();\ncellValue = _props.getFillValue();\n@@ -392,15 +394,14 @@ public class ReaderTextCSVParallel extends MatrixReader {\ncellValue = UtilFunctions.parseToDouble(part, _props.getNAStrings());\nif(cellValue != 0) {\n- a[index] = cellValue;\n+ avals[apos+j] = cellValue;\nnnz++;\n}\n- index++;\n}\n// sanity checks (number of columns, fill values)\nIOUtilFunctions.checkAndRaiseErrorCSVEmptyField(cellStr, _props.isFill(), noFillEmpty);\nIOUtilFunctions.checkAndRaiseErrorCSVNumColumns(_split.toString(), cellStr, parts, _cLen);\n- row++;\n+ _row++;\n}\nreturn nnz;\n}\n@@ -421,9 +422,9 @@ public class ReaderTextCSVParallel extends MatrixReader {\nfinal String cellStr = value.toString().trim();\nfinal String[] parts = IOUtilFunctions.split(cellStr, _props.getDelim());\n- col = 0;\n- sb.allocate(row);\n- SparseRow r = sb.get(row);\n+ _col = 0;\n+ sb.allocate(_row);\n+ SparseRow r = sb.get(_row);\nfor(String part : parts) {\npart = part.trim();\n@@ -436,17 +437,17 @@ public class ReaderTextCSVParallel extends MatrixReader {\n}\nif(cellValue != 0) {\n- r.append(col, cellValue);\n+ r.append(_col, cellValue);\nnnz++;\n}\n- col++;\n+ _col++;\n}\n// sanity checks (number of columns, fill values)\nIOUtilFunctions.checkAndRaiseErrorCSVEmptyField(cellStr, _props.isFill(), noFillEmpty);\nIOUtilFunctions.checkAndRaiseErrorCSVNumColumns(_split.toString(), cellStr, parts, _cLen);\n- row++;\n+ _row++;\n}\nreturn nnz;\n}\n@@ -463,11 +464,11 @@ public class ReaderTextCSVParallel extends MatrixReader {\ndouble cellValue = 0;\nboolean noFillEmpty = false;\nwhile(reader.next(key, value)) {\n- col = 0;\n+ _col = 0;\nfinal String cellStr = value.toString().trim();\nfinal String[] parts = IOUtilFunctions.split(cellStr, _props.getDelim());\n- sb.allocate(row);\n- SparseRow r = sb.get(row);\n+ sb.allocate(_row);\n+ SparseRow r = sb.get(_row);\nfor(String part : parts) {\npart = part.trim();\nif(part.isEmpty()) {\n@@ -479,17 +480,17 @@ public class ReaderTextCSVParallel extends MatrixReader {\n}\nif(cellValue != 0) {\n- r.append(col, cellValue);\n+ r.append(_col, cellValue);\nnnz++;\n}\n- col++;\n+ _col++;\n}\n// sanity checks (number of columns, fill values)\nIOUtilFunctions.checkAndRaiseErrorCSVEmptyField(cellStr, _props.isFill(), noFillEmpty);\nIOUtilFunctions.checkAndRaiseErrorCSVNumColumns(_split.toString(), cellStr, parts, _cLen);\n- row++;\n+ _row++;\n}\nreturn nnz;\n}\n@@ -506,25 +507,25 @@ public class ReaderTextCSVParallel extends MatrixReader {\nlong nnz = 0;\ndouble cellValue = 0;\nwhile(reader.next(key, value)) {\n- col = 0;\n+ _col = 0;\nfinal String cellStr = value.toString().trim();\nfinal String[] parts = IOUtilFunctions.split(cellStr, _props.getDelim());\n- sb.allocate(row);\n- SparseRow r = sb.get(row);\n+ sb.allocate(_row);\n+ SparseRow r = sb.get(_row);\nfor(String part : parts) {\nif(!part.isEmpty()) {\ncellValue = Double.parseDouble(part);\nif(cellValue != 0) {\n- r.append(col, cellValue);\n+ r.append(_col, cellValue);\nnnz++;\n}\n}\n- col++;\n+ _col++;\n}\nIOUtilFunctions.checkAndRaiseErrorCSVNumColumns(_split.toString(), cellStr, parts, _cLen);\n- row++;\n+ _row++;\n}\nreturn nnz;\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupFactory.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupFactory.java", "diff": "@@ -152,7 +152,7 @@ public final class ColGroupFactory {\n@Override\npublic Collection<AColGroup> call() {\n- ArrayList<AColGroup> res = new ArrayList<AColGroup>();\n+ ArrayList<AColGroup> res = new ArrayList<>();\nTmp tmpMap = new Tmp();\nfor(CompressedSizeInfoColGroup g : _groups)\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/dictionary/ADictionary.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/dictionary/ADictionary.java", "diff": "@@ -34,6 +34,8 @@ import org.apache.sysds.runtime.matrix.operators.ScalarOperator;\n*/\npublic abstract class ADictionary implements Serializable {\n+ private static final long serialVersionUID = 9118692576356558592L;\n+\nprotected static final Log LOG = LogFactory.getLog(ADictionary.class.getName());\n/**\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/dictionary/Dictionary.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/dictionary/Dictionary.java", "diff": "@@ -42,6 +42,8 @@ import org.apache.sysds.utils.MemoryEstimates;\n*/\npublic class Dictionary extends ADictionary {\n+ private static final long serialVersionUID = -6517136537249507753L;\n+\nprivate final double[] _values;\npublic Dictionary(double[] values) {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/dictionary/MatrixBlockDictionary.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/dictionary/MatrixBlockDictionary.java", "diff": "@@ -37,6 +37,8 @@ import org.apache.sysds.runtime.matrix.operators.ScalarOperator;\npublic class MatrixBlockDictionary extends ADictionary {\n+ private static final long serialVersionUID = 2535887782150955098L;\n+\nprivate MatrixBlock _data;\npublic MatrixBlockDictionary(MatrixBlock data) {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/dictionary/QDictionary.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/dictionary/QDictionary.java", "diff": "@@ -42,6 +42,8 @@ import org.apache.sysds.utils.MemoryEstimates;\n*/\npublic class QDictionary extends ADictionary {\n+ private static final long serialVersionUID = 2100501253343438897L;\n+\nprotected double _scale;\nprotected byte[] _values;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/mapping/AMapToData.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/mapping/AMapToData.java", "diff": "@@ -28,6 +28,8 @@ import org.apache.commons.logging.LogFactory;\npublic abstract class AMapToData implements Serializable {\n+ private static final long serialVersionUID = 100512759972844714L;\n+\nprotected static final Log LOG = LogFactory.getLog(AMapToData.class.getName());\nprivate int nUnique;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/mapping/MapToBit.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/mapping/MapToBit.java", "diff": "@@ -29,6 +29,8 @@ import org.apache.sysds.utils.MemoryEstimates;\npublic class MapToBit extends AMapToData {\n+ private static final long serialVersionUID = -8065234231282619923L;\n+\nprivate final BitSet _data;\nprivate final int _size;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/mapping/MapToByte.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/mapping/MapToByte.java", "diff": "@@ -29,6 +29,8 @@ import org.apache.sysds.utils.MemoryEstimates;\npublic class MapToByte extends AMapToData {\n+ private static final long serialVersionUID = -2498505439667351828L;\n+\nprivate final byte[] _data;\npublic MapToByte(int unique, int size) {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/mapping/MapToChar.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/mapping/MapToChar.java", "diff": "@@ -29,6 +29,8 @@ import org.apache.sysds.utils.MemoryEstimates;\npublic class MapToChar extends AMapToData {\n+ private static final long serialVersionUID = 6315708056775476541L;\n+\nprivate final char[] _data;\npublic MapToChar(int unique, int size) {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/mapping/MapToInt.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/mapping/MapToInt.java", "diff": "@@ -29,6 +29,8 @@ import org.apache.sysds.utils.MemoryEstimates;\npublic class MapToInt extends AMapToData {\n+ private static final long serialVersionUID = -5557070920888782274L;\n+\nprivate final int[] _data;\npublic MapToInt(int unique, int size) {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/offset/AOffset.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/offset/AOffset.java", "diff": "@@ -39,6 +39,7 @@ import org.apache.commons.logging.LogFactory;\n*/\npublic abstract class AOffset implements Serializable {\n+ private static final long serialVersionUID = -4143271285905723425L;\nprotected static final Log LOG = LogFactory.getLog(AOffset.class.getName());\nprotected SoftReference<Map<Integer, AIterator>> skipIterators;\n@@ -89,7 +90,7 @@ public abstract class AOffset implements Serializable {\nsk.put(row, it);\n}\nelse {\n- Map<Integer, AIterator> nsk = new HashMap<Integer, AIterator>();\n+ Map<Integer, AIterator> nsk = new HashMap<>();\nnsk.put(row, it.clone());\nskipIterators = new SoftReference<>(nsk);\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/offset/OffsetByte.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/offset/OffsetByte.java", "diff": "@@ -28,6 +28,8 @@ import org.apache.sysds.utils.MemoryEstimates;\npublic class OffsetByte extends AOffset {\n+ private static final long serialVersionUID = -4716104973912491790L;\n+\nprivate final static int maxV = 255;\nprivate final byte[] offsets;\nprivate final int offsetToFirst;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/offset/OffsetChar.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/offset/OffsetChar.java", "diff": "@@ -28,6 +28,8 @@ import org.apache.sysds.utils.MemoryEstimates;\npublic class OffsetChar extends AOffset {\n+ private static final long serialVersionUID = -1192266421395964882L;\n+\nprivate final static int maxV = (int) Character.MAX_VALUE;\nprivate final char[] offsets;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/matrix/data/LibMatrixMult.java", "new_path": "src/main/java/org/apache/sysds/runtime/matrix/data/LibMatrixMult.java", "diff": "@@ -1154,10 +1154,15 @@ public class LibMatrixMult\nfinal int blocksizeK = 32;\nfinal int blocksizeI = 32;\n+ int rl1 = pm2 ? 0 : rl;\n+ int ru1 = pm2 ? m : ru;\n+ int rl2 = pm2 ? rl : 0;\n+ int ru2 = pm2 ? ru : cd;\n+\n//blocked execution\n- for( int bi = rl; bi < ru; bi+=blocksizeI )\n- for( int bk = 0, bimin = Math.min(ru, bi+blocksizeI); bk < cd; bk+=blocksizeK ) {\n- int bkmin = Math.min(cd, bk+blocksizeK);\n+ for( int bi = rl1; bi < ru1; bi+=blocksizeI )\n+ for( int bk = rl2, bimin = Math.min(ru1, bi+blocksizeI); bk < ru2; bk+=blocksizeK ) {\n+ int bkmin = Math.min(ru2, bk+blocksizeK);\n//core sub block matrix multiplication\nfor(int i = bi; i < bimin; i++) {\ndouble[] avals = a.values(i), cvals = c.values(i);\n@@ -3883,7 +3888,7 @@ public class LibMatrixMult\ndouble jvmMem = InfrastructureAnalyzer.getLocalMaxMemory();\nreturn (m1.rlen==1 && LOW_LEVEL_OPTIMIZATION && m2.clen>1 && !(m1.isUltraSparse()||m2.isUltraSparse()))\n|| (m1.rlen<=16 && LOW_LEVEL_OPTIMIZATION && m2.clen>1 && m2.rlen > m1.rlen\n- && ( !m1.isUltraSparse() && !m2.sparse ) //dense-dense / sparse/dense\n+ && ( !m1.isUltraSparse() && !(m1.sparse & m2.sparse) ) //dense-dense / sparse-dense / dense-sparse\n&& (long)k * 8 * m1.rlen * m2.clen < Math.max(MEM_OVERHEAD_THRESHOLD,0.01*jvmMem) );\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/cost/ComputationCostEstimator.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/cost/ComputationCostEstimator.java", "diff": "@@ -92,18 +92,18 @@ public class ComputationCostEstimator implements ICostEstimate {\nprivate double leftMultCost(CompressedSizeInfoColGroup g) {\nfinal int nCols = g.getColumns().length;\n- // final double preAggregateCost = _nRows * 2.5;\n- final double preAggregateCost = _nRows * 1.5;\n- // final double preAggregateCost = _nRows * 0.2;\n+ final double preAggregateCost = _nRows;\nfinal int numberTuples = g.getNumVals();\nfinal double tupleSparsity = g.getTupleSparsity();\nfinal double postScalingCost = (nCols > 1 && tupleSparsity > 0.4) ? numberTuples * nCols : numberTuples *\nnCols * tupleSparsity;\n- if(numberTuples > 64000)\n- return preAggregateCost + postScalingCost * 2;\n-\n+ if(numberTuples < 64000)\nreturn preAggregateCost + postScalingCost;\n+ else\n+ // scale up cost worse if there is higher number of tuples.\n+ return preAggregateCost * (numberTuples / 6400) + postScalingCost * (numberTuples / 64000);\n+\n}\nprivate static double rightMultCost(CompressedSizeInfoColGroup g) {\n@@ -121,8 +121,6 @@ public class ComputationCostEstimator implements ICostEstimate {\n}\nprivate double overlappingDecompressionCost(CompressedSizeInfoColGroup g) {\n- // final int nVal = g.getNumVals();\n- // return nVal < 512 ? _nRows : _nRows * _nColsInMatrix * (nVal / 64000 + 1);\nreturn _nRows * 16 * (g.getNumVals() / 64000 + 1);\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupSDCZeros.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupSDCZeros.java", "diff": "@@ -28,6 +28,8 @@ import org.apache.commons.lang.NotImplementedException;\nimport org.apache.sysds.runtime.compress.colgroup.dictionary.ADictionary;\nimport org.apache.sysds.runtime.compress.colgroup.dictionary.Dictionary;\nimport org.apache.sysds.runtime.compress.colgroup.mapping.AMapToData;\n+import org.apache.sysds.runtime.compress.colgroup.mapping.MapToByte;\n+import org.apache.sysds.runtime.compress.colgroup.mapping.MapToChar;\nimport org.apache.sysds.runtime.compress.colgroup.mapping.MapToFactory;\nimport org.apache.sysds.runtime.compress.colgroup.offset.AIterator;\nimport org.apache.sysds.runtime.compress.colgroup.offset.AOffset;\n@@ -241,6 +243,16 @@ public class ColGroupSDCZeros extends ColGroupValue {\nfinal double[] preAV = preAgg.getDenseBlockValues();\nfinal int numVals = getNumValues();\n+ if(_data instanceof MapToByte) {\n+ preAggregateDenseByte(m, preAV, ((MapToByte) _data).getBytes(), rl, ru, cl, cu, _numRows, numVals,\n+ _indexes);\n+ }\n+ else if(_data instanceof MapToChar) {\n+ preAggregateDenseChar(m, preAV, ((MapToChar) _data).getChars(), rl, ru, cl, cu, _numRows, numVals,\n+ _indexes);\n+ }\n+ else {\n+ // multi row iterator.\nfinal AIterator itStart = _indexes.getIterator(cl);\nAIterator it = null;\nfor(int rowLeft = rl, offOut = 0; rowLeft < ru; rowLeft++, offOut += numVals) {\n@@ -254,6 +266,46 @@ public class ColGroupSDCZeros extends ColGroupValue {\nif(it != null && cu < m.getNumColumns())\n_indexes.cacheIterator(it, cu);\n}\n+ }\n+\n+ private static void preAggregateDenseByte(final MatrixBlock m, final double[] preAV, final byte[] d, final int rl,\n+ final int ru, final int cl, final int cu, final int nRow, final int nVal, AOffset indexes) {\n+ final double[] mV = m.getDenseBlockValues();\n+ // multi row iterator.\n+ final AIterator itStart = indexes.getIterator(cl);\n+ AIterator it = null;\n+ for(int rowLeft = rl, offOut = 0; rowLeft < ru; rowLeft++, offOut += preAV.length) {\n+ final int offLeft = rowLeft * nRow;\n+ it = itStart.clone();\n+ while(it.value() < cu && it.hasNext()) {\n+ int i = it.value();\n+ int index = d[it.getDataIndexAndIncrement()] & 0xFF;\n+ preAV[offOut + index] += mV[offLeft + i];\n+ }\n+ }\n+ if(it != null && cu < m.getNumColumns())\n+ indexes.cacheIterator(it, cu);\n+ }\n+\n+ private static void preAggregateDenseChar(final MatrixBlock m, final double[] preAV, final char[] d, final int rl,\n+ final int ru, final int cl, final int cu, final int nRow, final int nVal, AOffset indexes) {\n+ final double[] mV = m.getDenseBlockValues();\n+ // multi row iterator.\n+ final AIterator itStart = indexes.getIterator(cl);\n+ AIterator it = null;\n+ for(int rowLeft = rl, offOut = 0; rowLeft < ru; rowLeft++, offOut += preAV.length) {\n+ final int offLeft = rowLeft * nRow;\n+ it = itStart.clone();\n+ while(it.value() < cu && it.hasNext()) {\n+ int i = it.value();\n+ int index = d[it.getDataIndexAndIncrement()];\n+ preAV[offOut + index] += mV[offLeft + i];\n+ }\n+ }\n+ if(it != null && cu < m.getNumColumns())\n+ indexes.cacheIterator(it, cu);\n+\n+ }\nprivate void preAggregateDenseOld(MatrixBlock m, MatrixBlock preAgg, int rl, int ru) {\nfinal double[] preAV = preAgg.getDenseBlockValues();\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/paramserv/LocalDataPartitionerTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/paramserv/LocalDataPartitionerTest.java", "diff": "@@ -32,6 +32,7 @@ import org.apache.sysds.runtime.matrix.data.MatrixBlock;\nimport scala.Tuple2;\[email protected]\npublic class LocalDataPartitionerTest extends BaseDataPartitionerTest {\n@Test\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/paramserv/ParamservRecompilationTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/paramserv/ParamservRecompilationTest.java", "diff": "@@ -23,6 +23,7 @@ import org.junit.Test;\nimport org.apache.sysds.test.AutomatedTestBase;\nimport org.apache.sysds.test.TestConfiguration;\[email protected]\npublic class ParamservRecompilationTest extends AutomatedTestBase {\nprivate static final String TEST_NAME1 = \"paramserv-large-parallelism\";\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/paramserv/ParamservSyntaxTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/paramserv/ParamservSyntaxTest.java", "diff": "@@ -25,6 +25,7 @@ import org.apache.sysds.test.AutomatedTestBase;\nimport org.apache.sysds.test.TestConfiguration;\nimport org.junit.Test;\[email protected]\npublic class ParamservSyntaxTest extends AutomatedTestBase {\nprivate static final String TEST_NAME1 = \"paramserv-all-args\";\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/functions/paramserv/SparkDataPartitionerTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/paramserv/SparkDataPartitionerTest.java", "diff": "@@ -35,6 +35,7 @@ import org.apache.sysds.runtime.matrix.data.MatrixBlock;\nimport scala.Tuple2;\[email protected]\npublic class SparkDataPartitionerTest extends BaseDataPartitionerTest {\nprivate static SparkExecutionContext _sec;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/AColGroup.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/AColGroup.java", "diff": "@@ -516,6 +516,13 @@ public abstract class AColGroup implements Serializable {\n*/\npublic abstract AColGroup replace(double pattern, double replace);\n+ /**\n+ * Compute the column sum\n+ *\n+ * @param c The array to add the column sum to.\n+ */\n+ public abstract void computeColSums(double[] c);\n+\n@Override\npublic String toString() {\nStringBuilder sb = new StringBuilder();\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupCompressed.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupCompressed.java", "diff": "@@ -72,6 +72,10 @@ public abstract class ColGroupCompressed extends AColGroup {\nprotected abstract void computeRowSums(double[] c, boolean square, int rl, int ru);\n+ public void computeColSums(double[] c){\n+ computeColSums(c, false);\n+ }\n+\nprotected abstract void computeColSums(double[] c, boolean square);\nprotected abstract void computeRowMxx(double[] c, Builtin builtin, int rl, int ru);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupSDC.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupSDC.java", "diff": "@@ -507,11 +507,10 @@ public class ColGroupSDC extends ColGroupValue {\n@Override\npublic Dictionary preAggregateThatSDCZerosStructure(ColGroupSDCZeros that, Dictionary ret) {\n-\nfinal AIterator itThat = that._indexes.getIterator();\nfinal AIterator itThis = _indexes.getIterator();\nfinal int nCol = that._colIndexes.length;\n- final int defThis = this.getNumValues() * nCol - nCol;\n+ final int defThis = getNumValues() - 1;\nwhile(itThat.hasNext()) {\nfinal int thatV = itThat.value();\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupSDCZeros.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupSDCZeros.java", "diff": "@@ -436,7 +436,26 @@ public class ColGroupSDCZeros extends ColGroupValue {\n@Override\npublic Dictionary preAggregateThatSDCStructure(ColGroupSDC that, Dictionary ret, boolean preModified) {\n- throw new NotImplementedException();\n+ if(preModified){\n+ final AIterator itThat = that._indexes.getIterator();\n+ final AIterator itThis = _indexes.getIterator();\n+ final int nCol = that._colIndexes.length;\n+\n+ while(itThat.hasNext() && itThis.hasNext()) {\n+ if(itThat.value() == itThis.value()) {\n+ final int fr = that.getIndex(itThat.getDataIndexAndIncrement());\n+ final int to = getIndex(itThis.getDataIndexAndIncrement());\n+ that._dict.addToEntry(ret, fr, to, nCol);\n+ }\n+ else if(itThat.value() < itThis.value())\n+ itThat.next();\n+ else\n+ itThis.next();\n+ }\n+ return ret;\n+ }else{\n+ throw new NotImplementedException(\"Not implemented not PreModded preaggregate of SDC\");\n+ }\n}\n@Override\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupUncompressed.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/colgroup/ColGroupUncompressed.java", "diff": "@@ -620,4 +620,19 @@ public class ColGroupUncompressed extends AColGroup {\nMatrixBlock replaced = _data.replaceOperations(new MatrixBlock(), pattern, replace);\nreturn new ColGroupUncompressed(_colIndexes, replaced);\n}\n+\n+ @Override\n+ public void computeColSums(double[] c) {\n+ // TODO Auto-generated method stub\n+ MatrixBlock colSum = _data.colSum();\n+ if(colSum.isInSparseFormat()) {\n+ throw new NotImplementedException();\n+ }\n+ else {\n+ double[] dv = colSum.getDenseBlockValues();\n+ for(int i = 0; i < _colIndexes.length; i++)\n+ c[_colIndexes[i]] += dv[i];\n+\n+ }\n+ }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/cost/ComputationCostEstimator.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/cost/ComputationCostEstimator.java", "diff": "@@ -79,9 +79,12 @@ public class ComputationCostEstimator implements ICostEstimate {\ncost += _decompressions * decompressionCost(g);\ncost += _overlappingDecompressions * overlappingDecompressionCost(g);\n// 16 is assuming that the left side is 16 rows.\n- cost += _leftMultiplications * leftMultCost(g) * 16;\n+ double lmc = leftMultCost(g) * 16;\n+ cost += _leftMultiplications * lmc;\n// 16 is assuming that the right side is 16 rows.\n- cost += _rightMultiplications * rightMultCost(g) * 16;\n+ double rmc = rightMultCost(g) * 16;\n+ cost += _rightMultiplications * rmc;\n+ cost += _compressedMultiplication * (lmc + rmc);\ncost += _dictionaryOps * dictionaryOpsCost(g);\nreturn cost;\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/lib/CLALibLeftMultBy.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/lib/CLALibLeftMultBy.java", "diff": "@@ -91,54 +91,66 @@ public class CLALibLeftMultBy {\n}\npublic static void leftMultByTransposeSelf(CompressedMatrixBlock cmb, MatrixBlock result, int k) {\n- final int numColumns = cmb.getNumColumns();\nfinal boolean overlapping = cmb.isOverlapping();\n- List<AColGroup> groups = cmb.getColGroups();\n+ final List<AColGroup> groups = cmb.getColGroups();\n+\nresult.allocateDenseBlock();\nif(overlapping) {\nLOG.warn(\"Inefficient TSMM with overlapping matrix could be implemented multi-threaded but is not yet.\");\nleftMultByCompressedTransposedMatrix(groups, groups, result);\n}\n- else if(k <= 1) {\n- for(int i = 0; i < groups.size(); i++)\n- leftMultByCompressedTransposedMatrix(groups.get(i), groups, result, i, groups.size());\n- }\nelse {\n- try {\n- ExecutorService pool = CommonThreadPool.get(k);\n- ArrayList<Callable<Object>> tasks = new ArrayList<>();\n+ final boolean containsSDC = containsSDC(groups);\n+ final int numColumns = cmb.getNumColumns();\n+ final double[] constV = containsSDC ? new double[cmb.getNumColumns()] : null;\n+ final List<AColGroup> filteredGroups = filterSDCGroups(groups, constV);\n+ final double[] colSums = containsSDC ? new double[cmb.getNumColumns()] : null;\n+ if(containsSDC)\nfor(int i = 0; i < groups.size(); i++) {\n- final AColGroup g = groups.get(i);\n- tasks.add(new LeftMultByCompressedTransposedMatrixTask(groups, g, result, i, groups.size()));\n+ AColGroup gi = groups.get(i);\n+ if(!(gi instanceof ColGroupSDC || gi instanceof ColGroupSDCSingle))\n+ gi.computeColSums(colSums);\n}\n- for(Future<Object> tret : pool.invokeAll(tasks))\n- tret.get();\n- pool.shutdown();\n- }\n- catch(InterruptedException | ExecutionException e) {\n- throw new DMLRuntimeException(e);\n+ if(k <= 1)\n+ tsmmColGroups(groups, filteredGroups, result);\n+ else\n+ tsmmColGroupsParallel(groups, filteredGroups, result, k);\n+\n+ double[] retV = result.getDenseBlockValues();\n+\n+ // Move values in the lower part of the matrix to the upper part\n+ copyToUpperTriangle(retV, numColumns);\n+\n+ // add the correction layer for the subtracted common values.\n+ if(colSums != null) {\n+ outerProduct(colSums, constV, retV);\n+ addToUpperTriangle(retV, numColumns);\n}\n}\n- // Move values in the lower part of the matrix to the upper part\n- copyToUpperTriangle(result.getDenseBlockValues(), numColumns);\n- // calculate the number of non zeros, and allocate all value locations by copying upper triangle back to bottom.\n+\nlong nnz = LinearAlgebraUtils.copyUpperToLowerTriangle(result);\nresult.setNonZeros(nnz);\n- // Evaluate if the output should be sparsely allocated.\nresult.examSparsity();\n}\nprivate static void copyToUpperTriangle(final double[] c, final int cols) {\nfor(int i = 0, offC = 0; i < cols; i++, offC += cols)\n- for(int j = i, offR = i * cols; j < cols; j++, offR += cols) {\n+ for(int j = (i + 1), offR = (i + 1) * cols; j < cols; j++, offR += cols) {\nfinal double prev = c[offC + j];\nif(prev == 0)\nc[offC + j] = c[i + offR];\n+ c[i + offR] = 0;\n+ }\n}\n+ private static void addToUpperTriangle(final double[] c, final int cols) {\n+ for(int i = 0, offC = 0; i < cols; i++, offC += cols)\n+ for(int j = (i + 1), offR = (i + 1) * cols; j < cols; j++, offR += cols)\n+ c[offC + j] += c[i + offR];\n+\n}\nprivate static MatrixBlock leftMultByCompressedTransposedMatrix(List<AColGroup> colGroups,\n@@ -181,15 +193,6 @@ public class CLALibLeftMultBy {\nprivate final int _start;\nprivate final int _end;\n- protected LeftMultByCompressedTransposedMatrixTask(List<AColGroup> groups, AColGroup left, MatrixBlock ret,\n- int start, int end) {\n- _groups = groups;\n- _left = left;\n- _ret = ret;\n- _start = start;\n- _end = end;\n- }\n-\nprotected LeftMultByCompressedTransposedMatrixTask(List<AColGroup> groups, AColGroup left, MatrixBlock ret) {\n_groups = groups;\n_left = left;\n@@ -227,9 +230,85 @@ public class CLALibLeftMultBy {\nelse\nrhs.tsmm(ret);\n}\n+ }\n+ private static void tsmmColGroups(List<AColGroup> groups, List<AColGroup> filteredGroups, MatrixBlock ret) {\n+ for(int i = 0; i < groups.size(); i++)\n+ tsmmColGroupsIndexI(groups, filteredGroups, ret, i);\n}\n+ private static void tsmmColGroupsParallel(List<AColGroup> groups, List<AColGroup> filteredGroups, MatrixBlock ret,\n+ int k) {\n+ try {\n+ ExecutorService pool = CommonThreadPool.get(k);\n+ ArrayList<Callable<Object>> tasks = new ArrayList<>();\n+\n+ for(int i = 0; i < filteredGroups.size(); i++)\n+ tasks.add(new tsmmColGroupTask(groups, filteredGroups, ret, i));\n+\n+ for(Future<Object> tret : pool.invokeAll(tasks))\n+ tret.get();\n+ pool.shutdown();\n+ }\n+ catch(InterruptedException | ExecutionException e) {\n+ throw new DMLRuntimeException(e);\n+ }\n+ }\n+\n+ private static void tsmmColGroupsIndexI(List<AColGroup> groups, List<AColGroup> filteredGroups, MatrixBlock ret,\n+ int i) {\n+ final AColGroup full_lhs = groups.get(i);\n+ final AColGroup lhs = filteredGroups.get(i);\n+ final int start = i;\n+ final int end = groups.size();\n+ full_lhs.tsmm(ret);\n+ boolean isSDC = full_lhs instanceof ColGroupSDC || full_lhs instanceof ColGroupSDCSingle;\n+ // if(isSDC) {\n+ // Arrays.fill(tmp, 0);\n+ // full_lhs.computeColSums(tmp);\n+ // }\n+ for(int id = start + 1; id < end; id++) {\n+ final AColGroup full_rhs = groups.get(id);\n+ final AColGroup rhs = filteredGroups.get(id);\n+ if(isSDC && (full_rhs instanceof ColGroupSDC || full_rhs instanceof ColGroupSDCSingle)) {\n+ // Full\n+ full_lhs.leftMultByAColGroup(full_rhs, ret);\n+\n+ // Partial\n+ // full_lhs.leftMultByAColGroup(rhs, ret);\n+ // multiplyWithMostCommonElement(tmp, (ColGroupValue) full_rhs, ret);\n+ }\n+ else {\n+ lhs.leftMultByAColGroup(rhs, ret);\n+ }\n+ }\n+ }\n+\n+ // private static void multiplyWithMostCommonElement(double[] colSum, ColGroupValue full, MatrixBlock ret) {\n+ // final ADictionary d = full.getDictionary();\n+ // final double[] result = ret.getDenseBlockValues();\n+ // final int numVals = full.getNumValues();\n+ // final int[] colIndexes = full.getColIndices();\n+ // final int numColumns = ret.getNumColumns();\n+ // if(d instanceof MatrixBlockDictionary && ((MatrixBlockDictionary) d).getMatrixBlock().isInSparseFormat()) {\n+ // throw new NotImplementedException();\n+ // }\n+ // else {\n+ // final int offsetToDefault = numVals * full.getNumCols() - numVals;\n+ // final double[] dv = d.getValues();\n+ // for(int row = 0; row < colSum.length; row++) {\n+\n+ // final int offOut = numColumns * row;\n+ // final double vLeft = colSum[row];\n+ // if(vLeft != 0) {\n+ // for(int colId = 0; colId < colIndexes.length; colId++) {\n+ // result[offOut + colIndexes[colId]] += vLeft * dv[offsetToDefault + colId];\n+ // }\n+ // }\n+ // }\n+ // }\n+ // }\n+\nprivate static MatrixBlock leftMultByMatrix(List<AColGroup> colGroups, MatrixBlock that, MatrixBlock ret, int k,\nboolean overlapping) {\n@@ -237,28 +316,13 @@ public class CLALibLeftMultBy {\nret.setNonZeros(0);\nreturn ret;\n}\n- final int numColumnsOut = ret.getNumColumns();\n- boolean containsSDC = false;\n- for(AColGroup g : colGroups) {\n- if(g instanceof ColGroupSDC || g instanceof ColGroupSDCSingle)\n- containsSDC = true;\n- }\n+ final int numColumnsOut = ret.getNumColumns();\n+ final boolean containsSDC = containsSDC(colGroups);\n- final List<AColGroup> filteredGroups = containsSDC ? new ArrayList<>() : colGroups;\n// a constant colgroup summing the default values.\nfinal double[] constV = containsSDC ? new double[numColumnsOut] : null;\n-\n- if(containsSDC) {\n- for(AColGroup g : colGroups) {\n- if(g instanceof ColGroupSDC)\n- filteredGroups.add(((ColGroupSDC) g).extractCommon(constV));\n- else if(g instanceof ColGroupSDCSingle)\n- filteredGroups.add(((ColGroupSDCSingle) g).extractCommon(constV));\n- else\n- filteredGroups.add(g);\n- }\n- }\n+ final List<AColGroup> filteredGroups = filterSDCGroups(colGroups, constV);\nret.allocateDenseBlock();\nfinal double[] rowSums = containsSDC ? new double[that.getNumRows()] : null;\n@@ -418,6 +482,32 @@ public class CLALibLeftMultBy {\n}\n}\n+ private static class tsmmColGroupTask implements Callable<Object> {\n+ private final List<AColGroup> _groups;\n+ private final List<AColGroup> _filteredGroups;\n+ private final MatrixBlock _ret;\n+ private final int _index;\n+\n+ protected tsmmColGroupTask(List<AColGroup> groups, List<AColGroup> filteredGroups, MatrixBlock ret, int i) {\n+ _groups = groups;\n+ _filteredGroups = filteredGroups;\n+ _ret = ret;\n+ _index = i;\n+ }\n+\n+ @Override\n+ public MatrixBlock call() {\n+ try {\n+ tsmmColGroupsIndexI(_groups, _filteredGroups, _ret, _index);\n+ }\n+ catch(Exception e) {\n+ e.printStackTrace();\n+ throw new DMLRuntimeException(e);\n+ }\n+ return _ret;\n+ }\n+ }\n+\nprivate static void leftMultByMatrixPrimitive(List<AColGroup> colGroups, MatrixBlock that, MatrixBlock ret, int rl,\nint ru, double[] rowSums) {\nif(that.isInSparseFormat())\n@@ -538,4 +628,33 @@ public class CLALibLeftMultBy {\nCollections.sort(ColGroupValues, Comparator.comparing(AColGroup::getNumValues).reversed());\nreturn ColGroupValues;\n}\n+\n+ private static boolean containsSDC(List<AColGroup> groups) {\n+ boolean containsSDC = false;\n+\n+ for(AColGroup g : groups) {\n+ if(g instanceof ColGroupSDC || g instanceof ColGroupSDCSingle) {\n+ containsSDC = true;\n+ break;\n+ }\n+ }\n+ return containsSDC;\n+ }\n+\n+ private static List<AColGroup> filterSDCGroups(List<AColGroup> groups, double[] constV) {\n+ if(constV != null) {\n+ final List<AColGroup> filteredGroups = new ArrayList<>();\n+ for(AColGroup g : groups) {\n+ if(g instanceof ColGroupSDC)\n+ filteredGroups.add(((ColGroupSDC) g).extractCommon(constV));\n+ else if(g instanceof ColGroupSDCSingle)\n+ filteredGroups.add(((ColGroupSDCSingle) g).extractCommon(constV));\n+ else\n+ filteredGroups.add(g);\n+ }\n+ return filteredGroups;\n+ }\n+ else\n+ return groups;\n+ }\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/caching/CacheableData.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/caching/CacheableData.java", "diff": "@@ -789,7 +789,6 @@ public abstract class CacheableData<T extends CacheBlock> extends Data\n*/\npublic synchronized void exportData( int replication ) {\nexportData(_hdfsFileName, null, replication, null);\n- _hdfsFileExists = true;\n}\npublic synchronized void exportData(String fName, String outputFormat) {\n@@ -946,6 +945,7 @@ public abstract class CacheableData<T extends CacheBlock> extends Data\nLOG.trace(this.getDebugName() + \": Skip export to hdfs since data already exists.\");\n}\n+ _hdfsFileExists = true;\nif( DMLScript.STATISTICS ){\nlong t1 = System.nanoTime();\nCacheStatistics.incrementExportTime(t1-t0);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/caching/MatrixObject.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/caching/MatrixObject.java", "diff": "@@ -34,7 +34,6 @@ import org.apache.sysds.conf.ConfigurationManager;\nimport org.apache.sysds.hops.OptimizerUtils;\nimport org.apache.sysds.lops.Lop;\nimport org.apache.sysds.runtime.DMLRuntimeException;\n-import org.apache.sysds.runtime.compress.CompressedMatrixBlock;\nimport org.apache.sysds.runtime.controlprogram.ParForProgramBlock.PDataPartitionFormat;\nimport org.apache.sysds.runtime.controlprogram.context.SparkExecutionContext;\nimport org.apache.sysds.runtime.controlprogram.federated.FederatedRange;\n@@ -545,9 +544,6 @@ public class MatrixObject extends CacheableData<MatrixBlock> {\nReaderWriterFederated.write(fname, this._fedMapping);\n}\nelse if(_data != null) {\n- if(_data instanceof CompressedMatrixBlock)\n- _data = CompressedMatrixBlock.getUncompressed(_data);\n-\nMetaDataFormat iimd = (MetaDataFormat) _metaData;\n// Get the dimension information from the metadata stored within MatrixObject\nDataCharacteristics mc = iimd.getDataCharacteristics();\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/cost/ComputationCostEstimator.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/cost/ComputationCostEstimator.java", "diff": "@@ -84,7 +84,8 @@ public class ComputationCostEstimator implements ICostEstimate {\n// 16 is assuming that the right side is 16 rows.\ndouble rmc = rightMultCost(g) * 16;\ncost += _rightMultiplications * rmc;\n- cost += _compressedMultiplication * (lmc + rmc);\n+ // cost += _compressedMultiplication * (lmc + rmc);\n+ cost += _compressedMultiplication * _compressedMultCost(g);\ncost += _dictionaryOps * dictionaryOpsCost(g);\nreturn cost;\n}\n@@ -97,24 +98,38 @@ public class ComputationCostEstimator implements ICostEstimate {\nfinal int nCols = g.getColumns().length;\nfinal double preAggregateCost = _nRows;\n- final int numberTuples = g.getNumVals();\n+ final double numberTuples = g.getNumVals();\nfinal double tupleSparsity = g.getTupleSparsity();\n- final double postScalingCost = (nCols > 1 && tupleSparsity > 0.4) ? numberTuples * nCols : numberTuples *\n- nCols * tupleSparsity;\n+ final double postScalingCost = (nCols > 1 && tupleSparsity > 0.4) ? numberTuples * nCols * tupleSparsity *\n+ 1.4 : numberTuples * nCols;\nif(numberTuples < 64000)\nreturn preAggregateCost + postScalingCost;\nelse\n- // scale up cost worse if there is higher number of tuples.\nreturn preAggregateCost * (numberTuples / 6400) + postScalingCost * (numberTuples / 64000);\n}\n+ private double _compressedMultCost(CompressedSizeInfoColGroup g) {\n+ final int nCols = g.getColumns().length;\n+ final double mcf = g.getMostCommonFraction();\n+ final double preAggregateCost = mcf > 0.6 ? _nRows * (1 - 0.7 * mcf) : _nRows;\n+\n+ final double numberTuples = (float) g.getNumVals();\n+ final double tupleSparsity = g.getTupleSparsity();\n+ final double postScalingCost = (nCols > 1 && tupleSparsity > 0.4) ? numberTuples * nCols * tupleSparsity *\n+ 1.4 : numberTuples * nCols;\n+ if(numberTuples < 64000)\n+ return preAggregateCost + postScalingCost;\n+ else\n+ return preAggregateCost * (numberTuples / 64000) + postScalingCost * (numberTuples / 64000);\n+ }\n+\nprivate static double rightMultCost(CompressedSizeInfoColGroup g) {\nfinal int nCols = g.getColumns().length;\nfinal int numberTuples = g.getNumVals() * 10;\nfinal double tupleSparsity = g.getTupleSparsity();\n- final double postScalingCost = (nCols > 1 && tupleSparsity > 0.4) ? numberTuples * nCols : numberTuples *\n- nCols * tupleSparsity;\n+ final double postScalingCost = (nCols > 1 && tupleSparsity > 0.4) ? numberTuples * nCols * tupleSparsity *\n+ 1.4 : numberTuples * nCols;\nreturn postScalingCost;\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/compress/lib/CLALibLeftMultBy.java", "new_path": "src/main/java/org/apache/sysds/runtime/compress/lib/CLALibLeftMultBy.java", "diff": "@@ -233,18 +233,26 @@ public class CLALibLeftMultBy {\n}\nprivate static void tsmmColGroups(List<AColGroup> groups, List<AColGroup> filteredGroups, MatrixBlock ret) {\n- for(int i = 0; i < groups.size(); i++)\n+ for(int i = 0; i < groups.size(); i++) {\n+ groups.get(i).tsmm(ret);\ntsmmColGroupsIndexI(groups, filteredGroups, ret, i);\n}\n+ }\nprivate static void tsmmColGroupsParallel(List<AColGroup> groups, List<AColGroup> filteredGroups, MatrixBlock ret,\nint k) {\ntry {\nExecutorService pool = CommonThreadPool.get(k);\nArrayList<Callable<Object>> tasks = new ArrayList<>();\n+ // if(groups.size()< 10){\n- for(int i = 0; i < filteredGroups.size(); i++)\n- tasks.add(new tsmmColGroupTask(groups, filteredGroups, ret, i));\n+ // }\n+ final int numColGroups = groups.size();\n+ for(int i = 0; i < numColGroups; i++) {\n+ tasks.add(new tsmmSelfColGroupTask(groups.get(i), ret));\n+ for(int j = i +1; j < numColGroups; j++)\n+ tasks.add(new tsmmColGroupTask(groups, filteredGroups, ret, i, j, j+1));\n+ }\nfor(Future<Object> tret : pool.invokeAll(tasks))\ntret.get();\n@@ -257,57 +265,24 @@ public class CLALibLeftMultBy {\nprivate static void tsmmColGroupsIndexI(List<AColGroup> groups, List<AColGroup> filteredGroups, MatrixBlock ret,\nint i) {\n+ tsmmColGroupsIndexIStartEnd(groups, filteredGroups, ret, i, i + 1, groups.size());\n+ }\n+\n+ private static void tsmmColGroupsIndexIStartEnd(List<AColGroup> groups, List<AColGroup> filteredGroups,\n+ MatrixBlock ret, int i, int start, int end) {\nfinal AColGroup full_lhs = groups.get(i);\nfinal AColGroup lhs = filteredGroups.get(i);\n- final int start = i;\n- final int end = groups.size();\n- full_lhs.tsmm(ret);\nboolean isSDC = full_lhs instanceof ColGroupSDC || full_lhs instanceof ColGroupSDCSingle;\n- // if(isSDC) {\n- // Arrays.fill(tmp, 0);\n- // full_lhs.computeColSums(tmp);\n- // }\n- for(int id = start + 1; id < end; id++) {\n+ for(int id = start ; id < end; id++) {\nfinal AColGroup full_rhs = groups.get(id);\nfinal AColGroup rhs = filteredGroups.get(id);\n- if(isSDC && (full_rhs instanceof ColGroupSDC || full_rhs instanceof ColGroupSDCSingle)) {\n- // Full\n+ if(isSDC && (full_rhs instanceof ColGroupSDC || full_rhs instanceof ColGroupSDCSingle))\nfull_lhs.leftMultByAColGroup(full_rhs, ret);\n-\n- // Partial\n- // full_lhs.leftMultByAColGroup(rhs, ret);\n- // multiplyWithMostCommonElement(tmp, (ColGroupValue) full_rhs, ret);\n- }\n- else {\n+ else\nlhs.leftMultByAColGroup(rhs, ret);\n+\n}\n}\n- }\n-\n- // private static void multiplyWithMostCommonElement(double[] colSum, ColGroupValue full, MatrixBlock ret) {\n- // final ADictionary d = full.getDictionary();\n- // final double[] result = ret.getDenseBlockValues();\n- // final int numVals = full.getNumValues();\n- // final int[] colIndexes = full.getColIndices();\n- // final int numColumns = ret.getNumColumns();\n- // if(d instanceof MatrixBlockDictionary && ((MatrixBlockDictionary) d).getMatrixBlock().isInSparseFormat()) {\n- // throw new NotImplementedException();\n- // }\n- // else {\n- // final int offsetToDefault = numVals * full.getNumCols() - numVals;\n- // final double[] dv = d.getValues();\n- // for(int row = 0; row < colSum.length; row++) {\n-\n- // final int offOut = numColumns * row;\n- // final double vLeft = colSum[row];\n- // if(vLeft != 0) {\n- // for(int colId = 0; colId < colIndexes.length; colId++) {\n- // result[offOut + colIndexes[colId]] += vLeft * dv[offsetToDefault + colId];\n- // }\n- // }\n- // }\n- // }\n- // }\nprivate static MatrixBlock leftMultByMatrix(List<AColGroup> colGroups, MatrixBlock that, MatrixBlock ret, int k,\nboolean overlapping) {\n@@ -487,18 +462,44 @@ public class CLALibLeftMultBy {\nprivate final List<AColGroup> _filteredGroups;\nprivate final MatrixBlock _ret;\nprivate final int _index;\n+ private final int _start;\n+ private final int _end;\n- protected tsmmColGroupTask(List<AColGroup> groups, List<AColGroup> filteredGroups, MatrixBlock ret, int i) {\n+ protected tsmmColGroupTask(List<AColGroup> groups, List<AColGroup> filteredGroups, MatrixBlock ret, int i, int start, int end) {\n_groups = groups;\n_filteredGroups = filteredGroups;\n_ret = ret;\n_index = i;\n+ _start = start;\n+ _end = end;\n+ }\n+\n+ @Override\n+ public MatrixBlock call() {\n+ try {\n+ tsmmColGroupsIndexIStartEnd(_groups, _filteredGroups, _ret, _index, _start, _end);\n+ }\n+ catch(Exception e) {\n+ e.printStackTrace();\n+ throw new DMLRuntimeException(e);\n+ }\n+ return _ret;\n+ }\n+ }\n+\n+ private static class tsmmSelfColGroupTask implements Callable<Object> {\n+ private final AColGroup _g;\n+ private final MatrixBlock _ret;\n+\n+ protected tsmmSelfColGroupTask(AColGroup g, MatrixBlock ret) {\n+ _g = g;\n+ _ret = ret;\n}\n@Override\npublic MatrixBlock call() {\ntry {\n- tsmmColGroupsIndexI(_groups, _filteredGroups, _ret, _index);\n+ _g.tsmm(_ret);\n}\ncatch(Exception e) {\ne.printStackTrace();\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/hops/recompile/LiteralReplacement.java", "new_path": "src/main/java/org/apache/sysds/hops/recompile/LiteralReplacement.java", "diff": "@@ -47,6 +47,7 @@ import org.apache.sysds.runtime.instructions.cp.Data;\nimport org.apache.sysds.runtime.instructions.cp.ListObject;\nimport org.apache.sysds.runtime.instructions.cp.ScalarObject;\nimport org.apache.sysds.runtime.instructions.cp.ScalarObjectFactory;\n+import org.apache.sysds.runtime.lineage.LineageItem;\nimport org.apache.sysds.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysds.utils.Statistics;\n@@ -81,7 +82,7 @@ public class LiteralReplacement\nlit = (lit==null) ? replaceLiteralFullUnaryAggregateRightIndexing(c, vars) : lit;\nlit = (lit==null) ? replaceTReadMatrixFromList(c, ec) : lit;\nlit = (lit==null) ? replaceTReadMatrixFromListAppend(c, ec) : lit;\n- lit = (lit==null) ? replaceTReadMatrixLookupFromList(c, vars) : lit;\n+ lit = (lit==null) ? replaceTReadMatrixLookupFromList(c, ec) : lit;\nlit = (lit==null) ? replaceTReadScalarLookupFromList(c, vars) : lit;\n}\n@@ -385,8 +386,9 @@ public class LiteralReplacement\nreturn ret;\n}\n- private static DataOp replaceTReadMatrixLookupFromList( Hop c, LocalVariableMap vars ) {\n+ private static DataOp replaceTReadMatrixLookupFromList( Hop c, ExecutionContext ec) {\n//pattern: as.matrix(X[i:i]) or as.matrix(X['a','a']) with X being a list\n+ LocalVariableMap vars = ec.getVariables();\nDataOp ret = null;\nif( HopRewriteUtils.isUnary(c, OpOp1.CAST_AS_MATRIX)\n&& c.getInput().get(0) instanceof IndexingOp ) {\n@@ -402,7 +404,11 @@ public class LiteralReplacement\nLiteralOp lit = (LiteralOp) ix.getInput().get(1);\nMatrixObject mo = (MatrixObject) (!lit.getValueType().isNumeric() ?\nlist.slice(lit.getName()) : list.slice((int)lit.getLongValue()-1));\n+ LineageItem li = !lit.getValueType().isNumeric() ?\n+ list.getLineageItem(lit.getName()) : list.getLineageItem((int)lit.getLongValue()-1);\nvars.put(varname, mo);\n+ if (DMLScript.LINEAGE)\n+ ec.getLineage().set(varname, li);\nret = HopRewriteUtils.createTransientRead(varname, c);\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/DataGenCPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/DataGenCPInstruction.java", "diff": "@@ -528,6 +528,16 @@ public class DataGenCPInstruction extends UnaryCPInstruction {\ntmpInstStr = replaceNonLiteral(tmpInstStr, seq_incr, 7, ec);\nbreak;\n}\n+ case FRAMEINIT: {\n+ tmpInstStr = InstructionUtils.replaceOperandName(tmpInstStr);\n+ CPOperand frameInp = new CPOperand(frame_data, ValueType.STRING, DataType.SCALAR, true);\n+ tmpInstStr = InstructionUtils.replaceOperand(tmpInstStr, 2, frameInp.getLineageLiteral());\n+ tmpInstStr = replaceNonLiteral(tmpInstStr, rows, 3, ec);\n+ tmpInstStr = replaceNonLiteral(tmpInstStr, cols, 4, ec);\n+ CPOperand schemaInp = new CPOperand(schema, ValueType.STRING, DataType.SCALAR, true);\n+ tmpInstStr = !schema.equalsIgnoreCase(\"NULL\")\n+ ? InstructionUtils.replaceOperand(tmpInstStr, 5, schemaInp.getLineageLiteral()) : tmpInstStr;\n+ }\ncase TIME:\n// only opcode (time) is sufficient to compute from lineage.\nbreak;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/FunctionCallCPInstruction.java", "new_path": "src/main/java/org/apache/sysds/runtime/instructions/cp/FunctionCallCPInstruction.java", "diff": "@@ -313,8 +313,9 @@ public class FunctionCallCPInstruction extends CPInstruction {\nreturn reuse;\n}\n- private static String getCacheFunctionName(String fname, FunctionProgramBlock fpb) {\n- return !fpb.hasThreadID() ? fname :\n+ private String getCacheFunctionName(String fname, FunctionProgramBlock fpb) {\n+ String tmpFname = !fpb.hasThreadID() ? fname :\nfname.substring(0, fname.lastIndexOf(Lop.CP_CHILD_THREAD+fpb.getThreadID()));\n+ return DMLProgram.constructFunctionKey(_namespace, tmpFname);\n}\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/parser/ParForStatementBlock.java", "new_path": "src/main/java/org/apache/sysds/parser/ParForStatementBlock.java", "diff": "@@ -677,7 +677,7 @@ public class ParForStatementBlock extends ForStatementBlock\nfor(DataIdentifier write : datsUpdated) {\nif( !c._var.equals( write.getName() ) ) continue;\n- if( cdt != DataType.MATRIX && cdt != DataType.LIST ) {\n+ if( cdt != DataType.MATRIX && cdt != DataType.FRAME && cdt != DataType.LIST ) {\n//cannot infer type, need to exit (conservative approach)\nthrow new LanguageException(\"PARFOR loop dependency analysis: cannot check \"\n+ \"for dependencies due to unknown datatype of var '\"+c._var+\"': \"+cdt.name()+\".\");\n@@ -716,6 +716,7 @@ public class ParForStatementBlock extends ForStatementBlock\nreturn;\n}\nelse if( (cdt == DataType.MATRIX && dat2dt == DataType.MATRIX)\n+ || (cdt == DataType.FRAME && dat2dt == DataType.FRAME )\n|| (cdt == DataType.LIST && dat2dt == DataType.LIST ) )\n{\nboolean invalid = false;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/ParForProgramBlock.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/ParForProgramBlock.java", "diff": "@@ -38,6 +38,7 @@ import org.apache.sysds.parser.StatementBlock;\nimport org.apache.sysds.parser.VariableSet;\nimport org.apache.sysds.runtime.DMLRuntimeException;\nimport org.apache.sysds.runtime.controlprogram.caching.CacheableData;\n+import org.apache.sysds.runtime.controlprogram.caching.FrameObject;\nimport org.apache.sysds.runtime.controlprogram.caching.MatrixObject;\nimport org.apache.sysds.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysds.runtime.controlprogram.context.SparkExecutionContext;\n@@ -51,6 +52,7 @@ import org.apache.sysds.runtime.controlprogram.parfor.RemoteDPParForSpark;\nimport org.apache.sysds.runtime.controlprogram.parfor.RemoteParForJobReturn;\nimport org.apache.sysds.runtime.controlprogram.parfor.RemoteParForSpark;\nimport org.apache.sysds.runtime.controlprogram.parfor.ResultMerge;\n+import org.apache.sysds.runtime.controlprogram.parfor.ResultMergeFrameLocalMemory;\nimport org.apache.sysds.runtime.controlprogram.parfor.ResultMergeLocalAutomatic;\nimport org.apache.sysds.runtime.controlprogram.parfor.ResultMergeLocalFile;\nimport org.apache.sysds.runtime.controlprogram.parfor.ResultMergeLocalMemory;\n@@ -1056,9 +1058,9 @@ public class ParForProgramBlock extends ForProgramBlock\n* @param out output matrix\n* @param in array of input matrix objects\n*/\n- private static void cleanWorkerResultVariables(ExecutionContext ec, MatrixObject out, MatrixObject[] in, boolean parallel) {\n+ private static void cleanWorkerResultVariables(ExecutionContext ec, CacheableData<?> out, CacheableData<?>[] in, boolean parallel) {\n//check for empty inputs (no iterations executed)\n- Stream<MatrixObject> results = Arrays.stream(in).filter(m -> m!=null && m!=out);\n+ Stream<CacheableData<?>> results = Arrays.stream(in).filter(m -> m!=null && m!=out);\n//perform cleanup (parallel to mitigate file deletion bottlenecks)\n(parallel ? results.parallel() : results)\n.forEach(m -> ec.cleanupCacheableData(m));\n@@ -1307,34 +1309,42 @@ public class ParForProgramBlock extends ForProgramBlock\nreturn dp;\n}\n- private ResultMerge createResultMerge( PResultMerge prm, MatrixObject out, MatrixObject[] in, String fname, boolean accum, ExecutionContext ec )\n+ private ResultMerge<?> createResultMerge( PResultMerge prm,\n+ CacheableData<?> out, CacheableData<?>[] in, String fname, boolean accum, ExecutionContext ec )\n{\n- ResultMerge rm = null;\n+ ResultMerge<?> rm = null;\n+ if( out instanceof FrameObject ) {\n+ rm = new ResultMergeFrameLocalMemory((FrameObject)out, (FrameObject[])in, fname, accum);\n+ }\n+ else if(out instanceof MatrixObject) {\n//create result merge implementation (determine degree of parallelism\n//only for spark to avoid unnecessary spark context creation)\nswitch( prm )\n{\ncase LOCAL_MEM:\n- rm = new ResultMergeLocalMemory( out, in, fname, accum );\n+ rm = new ResultMergeLocalMemory( (MatrixObject)out, (MatrixObject[])in, fname, accum );\nbreak;\ncase LOCAL_FILE:\n- rm = new ResultMergeLocalFile( out, in, fname, accum );\n+ rm = new ResultMergeLocalFile( (MatrixObject)out, (MatrixObject[])in, fname, accum );\nbreak;\ncase LOCAL_AUTOMATIC:\n- rm = new ResultMergeLocalAutomatic( out, in, fname, accum );\n+ rm = new ResultMergeLocalAutomatic( (MatrixObject)out, (MatrixObject[])in, fname, accum );\nbreak;\ncase REMOTE_SPARK:\nint numMap = Math.max(_numThreads,\nSparkExecutionContext.getDefaultParallelism(true));\nint numRed = numMap; //equal map/reduce\n- rm = new ResultMergeRemoteSpark( out, in,\n- fname, accum, ec, numMap, numRed );\n+ rm = new ResultMergeRemoteSpark( (MatrixObject)out,\n+ (MatrixObject[])in, fname, accum, ec, numMap, numRed );\nbreak;\n-\ndefault:\nthrow new DMLRuntimeException(\"Undefined result merge: '\" +prm.toString()+\"'.\");\n}\n+ }\n+ else {\n+ throw new DMLRuntimeException(\"Unsupported result merge data: \"+out.getClass().getSimpleName());\n+ }\nreturn rm;\n}\n@@ -1437,14 +1447,15 @@ public class ParForProgramBlock extends ForProgramBlock\n{\nData dat = ec.getVariable(var._name);\n- if( dat instanceof MatrixObject ) //robustness scalars\n+ if( dat instanceof MatrixObject | dat instanceof FrameObject )\n{\n- MatrixObject out = (MatrixObject) dat;\n- MatrixObject[] in = Arrays.stream(results).map(vars ->\n- vars.get(var._name)).toArray(MatrixObject[]::new);\n+ CacheableData<?> out = (CacheableData<?>) dat;\n+ Stream<Object> tmp = Arrays.stream(results).map(vars -> vars.get(var._name));\n+ CacheableData<?>[] in = (dat instanceof MatrixObject) ?\n+ tmp.toArray(MatrixObject[]::new) : tmp.toArray(FrameObject[]::new);\nString fname = constructResultMergeFileName();\n- ResultMerge rm = createResultMerge(_resultMerge, out, in, fname, var._isAccum, ec);\n- MatrixObject outNew = USE_PARALLEL_RESULT_MERGE ?\n+ ResultMerge<?> rm = createResultMerge(_resultMerge, out, in, fname, var._isAccum, ec);\n+ CacheableData<?> outNew = USE_PARALLEL_RESULT_MERGE ?\nrm.executeParallelMerge(_numThreads) :\nrm.executeSerialMerge();\n@@ -1653,18 +1664,19 @@ public class ParForProgramBlock extends ForProgramBlock\nif( var == LocalTaskQueue.NO_MORE_TASKS ) // task queue closed (no more tasks)\nbreak;\n- MatrixObject out = null;\n+ CacheableData<?> out = null;\nsynchronized( _ec.getVariables() ){\n- out = _ec.getMatrixObject(var._name);\n+ out = _ec.getCacheableData(var._name);\n}\n- MatrixObject[] in = new MatrixObject[ _refVars.length ];\n- for( int i=0; i< _refVars.length; i++ )\n- in[i] = (MatrixObject) _refVars[i].get( var._name );\n+ Stream<Object> tmp = Arrays.stream(_refVars).map(vars -> vars.get(var._name));\n+ CacheableData<?>[] in = (out instanceof MatrixObject) ?\n+ tmp.toArray(MatrixObject[]::new) : tmp.toArray(FrameObject[]::new);\n+\nString fname = constructResultMergeFileName();\n- ResultMerge rm = createResultMerge(_resultMerge, out, in, fname, var._isAccum, _ec);\n- MatrixObject outNew = null;\n+ ResultMerge<?> rm = createResultMerge(_resultMerge, out, in, fname, var._isAccum, _ec);\n+ CacheableData<?> outNew = null;\nif( USE_PARALLEL_RESULT_MERGE )\noutNew = rm.executeParallelMerge( _numThreads );\nelse\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/caching/FrameObject.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/caching/FrameObject.java", "diff": "@@ -41,6 +41,7 @@ import org.apache.sysds.runtime.lineage.LineageItem;\nimport org.apache.sysds.runtime.lineage.LineageRecomputeUtils;\nimport org.apache.sysds.runtime.matrix.data.FrameBlock;\nimport org.apache.sysds.runtime.meta.DataCharacteristics;\n+import org.apache.sysds.runtime.meta.MatrixCharacteristics;\nimport org.apache.sysds.runtime.meta.MetaData;\nimport org.apache.sysds.runtime.meta.MetaDataFormat;\nimport org.apache.sysds.runtime.util.UtilFunctions;\n@@ -86,6 +87,12 @@ public class FrameObject extends CacheableData<FrameBlock>\n*/\npublic FrameObject(FrameObject fo) {\nsuper(fo);\n+\n+ MetaDataFormat metaOld = (MetaDataFormat) fo.getMetaData();\n+ _metaData = new MetaDataFormat(\n+ new MatrixCharacteristics(metaOld.getDataCharacteristics()),\n+ metaOld.getFileFormat());\n+ _schema = fo._schema.clone();\n}\n@Override\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/ResultMerge.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/ResultMerge.java", "diff": "@@ -21,34 +21,25 @@ package org.apache.sysds.runtime.controlprogram.parfor;\nimport org.apache.commons.logging.Log;\nimport org.apache.commons.logging.LogFactory;\n-import org.apache.sysds.runtime.controlprogram.caching.MatrixObject;\n-import org.apache.sysds.runtime.data.DenseBlock;\n+import org.apache.sysds.runtime.controlprogram.caching.CacheableData;\nimport org.apache.sysds.runtime.instructions.InstructionUtils;\n-import org.apache.sysds.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysds.runtime.matrix.operators.BinaryOperator;\nimport java.io.Serializable;\n-import java.util.List;\n-/**\n- * Due to independence of all iterations, any result has the following properties:\n- * (1) non local var, (2) matrix object, and (3) completely independent.\n- * These properties allow us to realize result merging in parallel without any synchronization.\n- *\n- */\n-public abstract class ResultMerge implements Serializable\n+public abstract class ResultMerge<T extends CacheableData<?>> implements Serializable\n{\n//note: this class needs to be serializable to ensure that all attributes of\n//ResultMergeRemoteSparkWCompare are included in the task closure\n- private static final long serialVersionUID = 2620430969346516677L;\n+ private static final long serialVersionUID = -6756689640511059030L;\nprotected static final Log LOG = LogFactory.getLog(ResultMerge.class.getName());\nprotected static final String NAME_SUFFIX = \"_rm\";\nprotected static final BinaryOperator PLUS = InstructionUtils.parseBinaryOperator(\"+\");\n//inputs to result merge\n- protected MatrixObject _output = null;\n- protected MatrixObject[] _inputs = null;\n+ protected T _output = null;\n+ protected T[] _inputs = null;\nprotected String _outputFName = null;\nprotected boolean _isAccum = false;\n@@ -56,7 +47,7 @@ public abstract class ResultMerge implements Serializable\n//do nothing\n}\n- public ResultMerge( MatrixObject out, MatrixObject[] in, String outputFilename, boolean accum ) {\n+ public ResultMerge( T out, T[] in, String outputFilename, boolean accum ) {\n_output = out;\n_inputs = in;\n_outputFName = outputFilename;\n@@ -70,7 +61,7 @@ public abstract class ResultMerge implements Serializable\n*\n* @return output (merged) matrix\n*/\n- public abstract MatrixObject executeSerialMerge();\n+ public abstract T executeSerialMerge();\n/**\n* Merge all given input matrices in parallel into the given output matrix.\n@@ -80,67 +71,6 @@ public abstract class ResultMerge implements Serializable\n* @param par degree of parallelism\n* @return output (merged) matrix\n*/\n- public abstract MatrixObject executeParallelMerge( int par );\n-\n- protected void mergeWithoutComp( MatrixBlock out, MatrixBlock in, boolean appendOnly ) {\n- mergeWithoutComp(out, in, appendOnly, false);\n- }\n-\n- protected void mergeWithoutComp( MatrixBlock out, MatrixBlock in, boolean appendOnly, boolean par ) {\n- //pass through to matrix block operations\n- if( _isAccum )\n- out.binaryOperationsInPlace(PLUS, in);\n- else\n- out.merge(in, appendOnly, par);\n- }\n-\n- /**\n- * NOTE: append only not applicable for wiht compare because output must be populated with\n- * initial state of matrix - with append, this would result in duplicates.\n- *\n- * @param out output matrix block\n- * @param in input matrix block\n- * @param compare ?\n- */\n- protected void mergeWithComp( MatrixBlock out, MatrixBlock in, DenseBlock compare )\n- {\n- //Notes for result correctness:\n- // * Always iterate over entire block in order to compare all values\n- // (using sparse iterator would miss values set to 0)\n- // * Explicit NaN awareness because for cases were original matrix contains\n- // NaNs, since NaN != NaN, otherwise we would potentially overwrite results\n- // * For the case of accumulation, we add out += (new-old) to ensure correct results\n- // because all inputs have the old values replicated\n-\n- if( in.isEmptyBlock(false) ) {\n- if( _isAccum ) return; //nothing to do\n- for( int i=0; i<in.getNumRows(); i++ )\n- for( int j=0; j<in.getNumColumns(); j++ )\n- if( compare.get(i, j) != 0 )\n- out.quickSetValue(i, j, 0);\n- }\n- else { //SPARSE/DENSE\n- int rows = in.getNumRows();\n- int cols = in.getNumColumns();\n- for( int i=0; i<rows; i++ )\n- for( int j=0; j<cols; j++ ) {\n- double valOld = compare.get(i,j);\n- double valNew = in.quickGetValue(i,j); //input value\n- if( (valNew != valOld && !Double.isNaN(valNew) ) //for changed values\n- || Double.isNaN(valNew) != Double.isNaN(valOld) ) //NaN awareness\n- {\n- double value = !_isAccum ? valNew :\n- (out.quickGetValue(i, j) + (valNew - valOld));\n- out.quickSetValue(i, j, value);\n- }\n- }\n- }\n- }\n+ public abstract T executeParallelMerge(int par);\n- protected long computeNonZeros( MatrixObject out, List<MatrixObject> in ) {\n- //sum of nnz of input (worker result) - output var existing nnz\n- long outNNZ = out.getDataCharacteristics().getNonZeros();\n- return outNNZ - in.size() * outNNZ + in.stream()\n- .mapToLong(m -> m.getDataCharacteristics().getNonZeros()).sum();\n- }\n}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/ResultMergeFrameLocalMemory.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysds.runtime.controlprogram.parfor;\n+\n+import org.apache.sysds.common.Types.ValueType;\n+import org.apache.sysds.runtime.DMLRuntimeException;\n+import org.apache.sysds.runtime.controlprogram.caching.FrameObject;\n+import org.apache.sysds.runtime.matrix.data.FrameBlock;\n+import org.apache.sysds.runtime.util.UtilFunctions;\n+\n+public class ResultMergeFrameLocalMemory extends ResultMerge<FrameObject>\n+{\n+ private static final long serialVersionUID = 549739254879310540L;\n+\n+ public ResultMergeFrameLocalMemory(FrameObject out, FrameObject[] in, String outputFilename, boolean accum) {\n+ super( out, in, outputFilename, accum );\n+ }\n+\n+ @Override\n+ public FrameObject executeSerialMerge()\n+ {\n+ FrameObject foNew = null; //always create new matrix object (required for nested parallelism)\n+\n+ if( LOG.isTraceEnabled() )\n+ LOG.trace(\"ResultMerge (local, in-memory): Execute serial merge for output \"\n+ +_output.hashCode()+\" (fname=\"+_output.getFileName()+\")\");\n+\n+ try\n+ {\n+ //get old and new output frame blocks\n+ FrameBlock outFB = _output.acquireRead();\n+ FrameBlock outFBNew = new FrameBlock(outFB);\n+\n+ //create compare matrix if required (existing data in result)\n+ FrameBlock compare = outFB;\n+ int rlen = compare.getNumRows();\n+ int clen = compare.getNumColumns();\n+\n+ //serial merge all inputs\n+ boolean flagMerged = false;\n+ for( FrameObject in : _inputs )\n+ {\n+ //check for empty inputs (no iterations executed)\n+ if( in != null && in != _output )\n+ {\n+ if( LOG.isTraceEnabled() )\n+ LOG.trace(\"ResultMergeFrame (local, in-memory): Merge input \"+in.hashCode()+\" (fname=\"+in.getFileName()+\")\");\n+\n+ //read/pin input_i\n+ FrameBlock inMB = in.acquireRead();\n+\n+ //core merge\n+ for(int j=0; j<clen; j++) {\n+ ValueType vt = compare.getSchema()[j];\n+ for(int i=0; i<rlen; i++) {\n+ Object val1 = compare.get(i, j);\n+ Object val2 = inMB.get(i, j);\n+ if( UtilFunctions.compareTo(vt, val1, val2) != 0 )\n+ outFBNew.set(i, j, val2);\n+ }\n+ }\n+\n+ //unpin and clear in-memory input_i\n+ in.release();\n+ in.clearData();\n+ flagMerged = true;\n+ }\n+ }\n+\n+ //create output and release old output\n+ foNew = flagMerged ? createNewFrameObject(_output, outFBNew) : _output;\n+ _output.release();\n+ }\n+ catch(Exception ex) {\n+ throw new DMLRuntimeException(ex);\n+ }\n+\n+ //LOG.trace(\"ResultMerge (local, in-memory): Executed serial merge for output \"+_output.getVarName()+\" (fname=\"+_output.getFileName()+\") in \"+time.stop()+\"ms\");\n+\n+ return foNew;\n+ }\n+\n+ @Override\n+ public FrameObject executeParallelMerge( int par ) {\n+ if( LOG.isTraceEnabled() )\n+ LOG.trace(\"ResultMerge (local, in-memory): Execute parallel (par=\"+par+\") \"\n+ + \"merge for output \"+_output.hashCode()+\" (fname=\"+_output.getFileName()+\")\");\n+ return executeSerialMerge();\n+ }\n+\n+ private static FrameObject createNewFrameObject( FrameObject foOld, FrameBlock dataNew ) {\n+ FrameObject ret = new FrameObject(foOld);\n+ ret.acquireModify(dataNew);\n+ ret.release();\n+ return ret;\n+ }\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/ResultMergeLocalAutomatic.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/ResultMergeLocalAutomatic.java", "diff": "@@ -26,11 +26,11 @@ import org.apache.sysds.runtime.controlprogram.parfor.opt.OptimizerRuleBased;\nimport org.apache.sysds.runtime.controlprogram.parfor.stat.Timing;\nimport org.apache.sysds.runtime.meta.DataCharacteristics;\n-public class ResultMergeLocalAutomatic extends ResultMerge\n+public class ResultMergeLocalAutomatic extends ResultMergeMatrix\n{\nprivate static final long serialVersionUID = 1600893100602101732L;\n- private ResultMerge _rm = null;\n+ private ResultMergeMatrix _rm = null;\npublic ResultMergeLocalAutomatic( MatrixObject out, MatrixObject[] in, String outputFilename, boolean accum ) {\nsuper( out, in, outputFilename, accum );\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/ResultMergeLocalFile.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/ResultMergeLocalFile.java", "diff": "@@ -67,7 +67,7 @@ import java.util.Map.Entry;\n* NOTE: file merge typically used due to memory constraints - parallel merge would increase the memory\n* consumption again.\n*/\n-public class ResultMergeLocalFile extends ResultMerge\n+public class ResultMergeLocalFile extends ResultMergeMatrix\n{\nprivate static final long serialVersionUID = -6905893742840020489L;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/ResultMergeLocalMemory.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/ResultMergeLocalMemory.java", "diff": "@@ -39,7 +39,7 @@ import java.util.ArrayList;\n*\n*\n*/\n-public class ResultMergeLocalMemory extends ResultMerge\n+public class ResultMergeLocalMemory extends ResultMergeMatrix\n{\nprivate static final long serialVersionUID = -3543612508601511701L;\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/ResultMergeMatrix.java", "diff": "+/*\n+ * Licensed to the Apache Software Foundation (ASF) under one\n+ * or more contributor license agreements. See the NOTICE file\n+ * distributed with this work for additional information\n+ * regarding copyright ownership. The ASF licenses this file\n+ * to you under the Apache License, Version 2.0 (the\n+ * \"License\"); you may not use this file except in compliance\n+ * with the License. You may obtain a copy of the License at\n+ *\n+ * http://www.apache.org/licenses/LICENSE-2.0\n+ *\n+ * Unless required by applicable law or agreed to in writing,\n+ * software distributed under the License is distributed on an\n+ * \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+ * KIND, either express or implied. See the License for the\n+ * specific language governing permissions and limitations\n+ * under the License.\n+ */\n+\n+package org.apache.sysds.runtime.controlprogram.parfor;\n+\n+import org.apache.sysds.runtime.controlprogram.caching.MatrixObject;\n+import org.apache.sysds.runtime.data.DenseBlock;\n+import org.apache.sysds.runtime.matrix.data.MatrixBlock;\n+\n+import java.io.Serializable;\n+import java.util.List;\n+\n+/**\n+ * Due to independence of all iterations, any result has the following properties:\n+ * (1) non local var, (2) matrix object, and (3) completely independent.\n+ * These properties allow us to realize result merging in parallel without any synchronization.\n+ *\n+ */\n+public abstract class ResultMergeMatrix extends ResultMerge<MatrixObject> implements Serializable\n+{\n+ private static final long serialVersionUID = 5319002218804570071L;\n+\n+ public ResultMergeMatrix() {\n+ super();\n+ }\n+\n+ public ResultMergeMatrix(MatrixObject out, MatrixObject[] in, String outputFilename, boolean accum) {\n+ super(out, in, outputFilename, accum);\n+ }\n+\n+ protected void mergeWithoutComp( MatrixBlock out, MatrixBlock in, boolean appendOnly ) {\n+ mergeWithoutComp(out, in, appendOnly, false);\n+ }\n+\n+ protected void mergeWithoutComp( MatrixBlock out, MatrixBlock in, boolean appendOnly, boolean par ) {\n+ //pass through to matrix block operations\n+ if( _isAccum )\n+ out.binaryOperationsInPlace(PLUS, in);\n+ else\n+ out.merge(in, appendOnly, par);\n+ }\n+\n+ /**\n+ * NOTE: append only not applicable for wiht compare because output must be populated with\n+ * initial state of matrix - with append, this would result in duplicates.\n+ *\n+ * @param out output matrix block\n+ * @param in input matrix block\n+ * @param compare ?\n+ */\n+ protected void mergeWithComp( MatrixBlock out, MatrixBlock in, DenseBlock compare )\n+ {\n+ //Notes for result correctness:\n+ // * Always iterate over entire block in order to compare all values\n+ // (using sparse iterator would miss values set to 0)\n+ // * Explicit NaN awareness because for cases were original matrix contains\n+ // NaNs, since NaN != NaN, otherwise we would potentially overwrite results\n+ // * For the case of accumulation, we add out += (new-old) to ensure correct results\n+ // because all inputs have the old values replicated\n+\n+ if( in.isEmptyBlock(false) ) {\n+ if( _isAccum ) return; //nothing to do\n+ for( int i=0; i<in.getNumRows(); i++ )\n+ for( int j=0; j<in.getNumColumns(); j++ )\n+ if( compare.get(i, j) != 0 )\n+ out.quickSetValue(i, j, 0);\n+ }\n+ else { //SPARSE/DENSE\n+ int rows = in.getNumRows();\n+ int cols = in.getNumColumns();\n+ for( int i=0; i<rows; i++ )\n+ for( int j=0; j<cols; j++ ) {\n+ double valOld = compare.get(i,j);\n+ double valNew = in.quickGetValue(i,j); //input value\n+ if( (valNew != valOld && !Double.isNaN(valNew) ) //for changed values\n+ || Double.isNaN(valNew) != Double.isNaN(valOld) ) //NaN awareness\n+ {\n+ double value = !_isAccum ? valNew :\n+ (out.quickGetValue(i, j) + (valNew - valOld));\n+ out.quickSetValue(i, j, value);\n+ }\n+ }\n+ }\n+ }\n+\n+ protected long computeNonZeros( MatrixObject out, List<MatrixObject> in ) {\n+ //sum of nnz of input (worker result) - output var existing nnz\n+ long outNNZ = out.getDataCharacteristics().getNonZeros();\n+ return outNNZ - in.size() * outNNZ + in.stream()\n+ .mapToLong(m -> m.getDataCharacteristics().getNonZeros()).sum();\n+ }\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/ResultMergeRemoteSpark.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/ResultMergeRemoteSpark.java", "diff": "@@ -44,7 +44,7 @@ import org.apache.sysds.utils.Statistics;\nimport java.util.Arrays;\n-public class ResultMergeRemoteSpark extends ResultMerge\n+public class ResultMergeRemoteSpark extends ResultMergeMatrix\n{\nprivate static final long serialVersionUID = -6924566953903424820L;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/ResultMergeRemoteSparkWCompare.java", "new_path": "src/main/java/org/apache/sysds/runtime/controlprogram/parfor/ResultMergeRemoteSparkWCompare.java", "diff": "@@ -31,7 +31,7 @@ import org.apache.sysds.runtime.util.DataConverter;\nimport scala.Tuple2;\n-public class ResultMergeRemoteSparkWCompare extends ResultMerge implements PairFunction<Tuple2<MatrixIndexes,Tuple2<Iterable<MatrixBlock>,MatrixBlock>>, MatrixIndexes, MatrixBlock>\n+public class ResultMergeRemoteSparkWCompare extends ResultMergeMatrix implements PairFunction<Tuple2<MatrixIndexes,Tuple2<Iterable<MatrixBlock>,MatrixBlock>>, MatrixIndexes, MatrixBlock>\n{\nprivate static final long serialVersionUID = -5970805069405942836L;\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysds/test/component/parfor/ParForDependencyAnalysisTest.java", "new_path": "src/test/java/org/apache/sysds/test/component/parfor/ParForDependencyAnalysisTest.java", "diff": "@@ -66,8 +66,8 @@ import org.apache.sysds.test.TestConfiguration;\n* 49a: dep, 49b: dep\n* * accumulators\n* 53a: no, 53b dep, 53c dep, 53d dep, 53e dep\n- * * lists\n- * 54a: no, 54b: no, 54c: dep, 54d: dep\n+ * * lists/frames\n+ * 54a: no, 54b: no, 54c: dep, 54d: dep, 54e: no-dep, 54f: dep\n* * negative loop increment\n* 55a: no, 55b: yes\n*/\n@@ -327,6 +327,12 @@ public class ParForDependencyAnalysisTest extends AutomatedTestBase\n@Test\npublic void testDependencyAnalysis54d() { runTest(\"parfor54d.dml\", true); }\n+ @Test\n+ public void testDependencyAnalysis54e() { runTest(\"parfor54e.dml\", false); }\n+\n+ @Test\n+ public void testDependencyAnalysis54f() { runTest(\"parfor54f.dml\", true); }\n+\n@Test\npublic void testDependencyAnalysis55a() { runTest(\"parfor55a.dml\", false); }\n" }, { "change_type": "RENAME", "old_path": "src/test/java/org/apache/sysds/test/functions/parfor/misc/ParForListResultVarsTest.java", "new_path": "src/test/java/org/apache/sysds/test/functions/parfor/misc/ParForListFrameResultVarsTest.java", "diff": "@@ -25,16 +25,18 @@ import org.apache.sysds.runtime.matrix.data.MatrixValue.CellIndex;\nimport org.apache.sysds.test.AutomatedTestBase;\nimport org.apache.sysds.test.TestConfiguration;\n-public class ParForListResultVarsTest extends AutomatedTestBase\n+public class ParForListFrameResultVarsTest extends AutomatedTestBase\n{\nprivate final static String TEST_DIR = \"functions/parfor/\";\nprivate final static String TEST_NAME1 = \"parfor_listResults\";\n- private final static String TEST_CLASS_DIR = TEST_DIR + ParForListResultVarsTest.class.getSimpleName() + \"/\";\n+ private final static String TEST_NAME2 = \"parfor_frameResults\";\n+\n+ private final static String TEST_CLASS_DIR = TEST_DIR + ParForListFrameResultVarsTest.class.getSimpleName() + \"/\";\n@Override\npublic void setUp() {\n- addTestConfiguration(TEST_NAME1,\n- new TestConfiguration(TEST_CLASS_DIR, TEST_NAME1, new String[] { \"R\" }) );\n+ addTestConfiguration(TEST_NAME1, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME1, new String[]{\"R\"}));\n+ addTestConfiguration(TEST_NAME2, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME2, new String[]{\"R\"}));\n}\n@Test\n@@ -47,11 +49,21 @@ public class ParForListResultVarsTest extends AutomatedTestBase\nrunListResultVarTest(TEST_NAME1, 35, 10);\n}\n+ @Test\n+ public void testParForFrameResult1a() {\n+ runListResultVarTest(TEST_NAME2, 2, 1);\n+ }\n+\n+ @Test\n+ public void testParForFrameResult1b() {\n+ runListResultVarTest(TEST_NAME2, 35, 10);\n+ }\n+\nprivate void runListResultVarTest(String testName, int rows, int cols) {\nloadTestConfiguration(getTestConfiguration(testName));\nString HOME = SCRIPT_DIR + TEST_DIR;\n- fullDMLScriptName = HOME + TEST_NAME1 + \".dml\";\n+ fullDMLScriptName = HOME + testName + \".dml\";\nprogramArgs = new String[]{\"-explain\",\"-args\",\nString.valueOf(rows), String.valueOf(cols), output(\"R\") };\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/component/parfor/parfor54e.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+\n+A = rbind(as.frame(\"a\"), as.frame(\"b\"), as.frame(\"c\"));\n+parfor( i in 1:nrow(A) )\n+ A[i,1] = as.frame(as.scalar(A[i,1])+\"-\"+i);\n+print(toString(A));\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/component/parfor/parfor54f.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+\n+A = rbind(as.frame(\"a\"), as.frame(\"b\"), as.frame(\"c\"));\n+parfor( i in 1:nrow(A) )\n+ A[i,1] = as.frame(as.scalar(A[1,1])+\"-\"+i);\n+print(toString(A));\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/parfor/parfor_frameResults.dml", "diff": "+#-------------------------------------------------------------\n+#\n+# Licensed to the Apache Software Foundation (ASF) under one\n+# or more contributor license agreements. See the NOTICE file\n+# distributed with this work for additional information\n+# regarding copyright ownership. The ASF licenses this file\n+# to you under the Apache License, Version 2.0 (the\n+# \"License\"); you may not use this file except in compliance\n+# with the License. You may obtain a copy of the License at\n+#\n+# http://www.apache.org/licenses/LICENSE-2.0\n+#\n+# Unless required by applicable law or agreed to in writing,\n+# software distributed under the License is distributed on an\n+# \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY\n+# KIND, either express or implied. See the License for the\n+# specific language governing permissions and limitations\n+# under the License.\n+#\n+#-------------------------------------------------------------\n+\n+F = as.frame(matrix(0,7,1));\n+\n+parfor(i in 1:nrow(F))\n+ F[i,1] = as.frame(rowMeans(as.matrix(F[i]))+i);\n+\n+R1 = matrix(0,0,1)\n+for(i in 1:length(F))\n+ R1 = rbind(R1, as.matrix(F[i,1]));\n+\n+R = as.matrix(sum(R1==seq(1,7)));\n+write(R, $3);\n" } ]