Datasets:

---

JetBrains-Research

/

commit-chronicle

like 9

Follow

JetBrains Research 70

[ { "change_type": "MODIFY", "old_path": "docs/python-performance-test.md", "new_path": "docs/python-performance-test.md", "diff": "@@ -177,7 +177,20 @@ In the example above `--tag` can be a major/minor systemml version and `--auth`\nCurrently we only support time difference between algorithms in different versions. This can be obtained by running the script below\n`./stats.py --auth client_json.json --exec-mode singlenode --tags 1.0 2.0`\n-Note: Please pip install `https://github.com/burnash/gspread` to use google docs client.\n+We pass different `matrix shapes` using `--mat-shape` argument.\n+\n+Matrix Shape | Approximate Data Size\n+--- | --- |\n+10k_1k|80MB\n+100k_1k|800MB\n+1M_1k|8GB\n+10M_1k|80GB\n+100M_1k|800GB\n+\n+For example the command below runs performance test for all data sizes described above\n+`run_perftest.py --family binomial clustering multinomial regression1 regression2 stats1 stats2 --mat-shape 10k_1k 100k_1k 1M_1k 10M_1k 100M_1k --master yarn-client --temp-dir hdfs://localhost:9000/user/systemml`\n+\n+Note: Please use this command `pip3 install -r requirements.txt` before using the perftest scripts.\n## Troubleshooting\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/utils_fs.py", "new_path": "scripts/perftest/python/utils_fs.py", "diff": "@@ -134,12 +134,12 @@ def relevant_folders(path, algo, family, matrix_type, matrix_shape, mode):\nif mode == 'data-gen':\nsub_folder_name = '.'.join([family, current_matrix_type, current_matrix_shape])\ncmd = ['hdfs', 'dfs', '-ls', path]\n- path_subdir = subprocess_exec(' '.join(cmd), 'dir')\n+ path_subdir = subprocess_exec(' '.join(cmd), extract='dir')\nif mode == 'train':\nsub_folder_name = '.'.join([algo, family, current_matrix_type, current_matrix_shape])\ncmd = ['hdfs', 'dfs', '-ls', path]\n- path_subdir = subprocess_exec(' '.join(cmd), 'dir')\n+ path_subdir = subprocess_exec(' '.join(cmd), extract='dir')\npath_folders = list(filter(lambda x: contains_dir(x, sub_folder_name), path_subdir))\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/OptimizerUtils.java", "new_path": "src/main/java/org/apache/sysml/hops/OptimizerUtils.java", "diff": "@@ -24,6 +24,8 @@ import java.util.HashMap;\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.sysml.api.DMLScript;\nimport org.apache.sysml.api.DMLScript.RUNTIME_PLATFORM;\nimport org.apache.sysml.conf.CompilerConfig;\n@@ -933,6 +935,11 @@ public class OptimizerUtils\nreturn ret;\n}\n+ public static Level getDefaultLogLevel() {\n+ Level log = Logger.getRootLogger().getLevel();\n+ return (log != null) ? log : Level.INFO;\n+ }\n+\n////////////////////////\n// Sparsity Estimates //\n////////////////////////\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/ParForStatementBlock.java", "new_path": "src/main/java/org/apache/sysml/parser/ParForStatementBlock.java", "diff": "@@ -128,7 +128,7 @@ public class ParForStatementBlock extends ForStatementBlock\n_paramDefaults.put( EXEC_MODE, String.valueOf(PExecMode.LOCAL) );\n_paramDefaults.put( OPT_MODE, String.valueOf(POptMode.RULEBASED) );\n_paramDefaults.put( PROFILE, \"0\" );\n- _paramDefaults.put( OPT_LOG, Logger.getRootLogger().getLevel().toString() );\n+ _paramDefaults.put( OPT_LOG, OptimizerUtils.getDefaultLogLevel().toString() );\n_paramDefaults2 = new HashMap<String, String>(); //OPT_MODE always specified\n_paramDefaults2.put( CHECK, \"1\" );\n@@ -139,7 +139,7 @@ public class ParForStatementBlock extends ForStatementBlock\n_paramDefaults2.put( RESULT_MERGE, String.valueOf(PResultMerge.UNSPECIFIED) );\n_paramDefaults2.put( EXEC_MODE, String.valueOf(PExecMode.UNSPECIFIED) );\n_paramDefaults2.put( PROFILE, \"0\" );\n- _paramDefaults2.put( OPT_LOG, Logger.getRootLogger().getLevel().toString() );\n+ _paramDefaults2.put( OPT_LOG, OptimizerUtils.getDefaultLogLevel().toString() );\n_idSeq = new IDSequence();\n_idSeqfn = new IDSequence();\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/ParameterizedBuiltinFunctionExpression.java", "new_path": "src/main/java/org/apache/sysml/parser/ParameterizedBuiltinFunctionExpression.java", "diff": "@@ -26,6 +26,7 @@ import java.util.HashSet;\nimport org.apache.sysml.hops.Hop.ParamBuiltinOp;\nimport org.apache.sysml.parser.LanguageException.LanguageErrorCodes;\n+import org.apache.wink.json4j.JSONObject;\npublic class ParameterizedBuiltinFunctionExpression extends DataIdentifier\n@@ -288,6 +289,7 @@ public class ParameterizedBuiltinFunctionExpression extends DataIdentifier\n//validate specification\ncheckDataValueType(\"transformapply\", TF_FN_PARAM_SPEC, DataType.SCALAR, ValueType.STRING, conditional);\n+ validateTransformSpec(TF_FN_PARAM_SPEC, conditional);\n//set output dimensions\noutput.setDataType(DataType.MATRIX);\n@@ -304,6 +306,7 @@ public class ParameterizedBuiltinFunctionExpression extends DataIdentifier\n//validate specification\ncheckDataValueType(\"transformdecode\", TF_FN_PARAM_SPEC, DataType.SCALAR, ValueType.STRING, conditional);\n+ validateTransformSpec(TF_FN_PARAM_SPEC, conditional);\n//set output dimensions\noutput.setDataType(DataType.FRAME);\n@@ -316,6 +319,7 @@ public class ParameterizedBuiltinFunctionExpression extends DataIdentifier\n{\n//validate specification\ncheckDataValueType(\"transformmeta\", TF_FN_PARAM_SPEC, DataType.SCALAR, ValueType.STRING, conditional);\n+ validateTransformSpec(TF_FN_PARAM_SPEC, conditional);\n//validate meta data path\ncheckDataValueType(\"transformmeta\", TF_FN_PARAM_MTD, DataType.SCALAR, ValueType.STRING, conditional);\n@@ -334,6 +338,7 @@ public class ParameterizedBuiltinFunctionExpression extends DataIdentifier\n//validate specification\ncheckDataValueType(\"transformencode\", TF_FN_PARAM_SPEC, DataType.SCALAR, ValueType.STRING, conditional);\n+ validateTransformSpec(TF_FN_PARAM_SPEC, conditional);\n//set output dimensions\noutput1.setDataType(DataType.MATRIX);\n@@ -344,6 +349,20 @@ public class ParameterizedBuiltinFunctionExpression extends DataIdentifier\noutput2.setDimensions(-1, -1);\n}\n+ private void validateTransformSpec(String pname, boolean conditional) throws LanguageException {\n+ Expression data = getVarParam(pname);\n+ if( data instanceof StringIdentifier ) {\n+ try {\n+ StringIdentifier spec = (StringIdentifier)data;\n+ new JSONObject(spec.getValue());\n+ }\n+ catch(Exception ex) {\n+ raiseValidateError(\"Transform specification parsing issue: \",\n+ conditional, ex.getMessage());\n+ }\n+ }\n+ }\n+\nprivate void validateReplace(DataIdentifier output, boolean conditional) throws LanguageException {\n//check existence and correctness of arguments\nExpression target = getVarParam(\"target\");\n@@ -721,11 +740,6 @@ public class ParameterizedBuiltinFunctionExpression extends DataIdentifier\n@Override\npublic boolean multipleReturns() {\n- switch(_opcode) {\n- case TRANSFORMENCODE:\n- return true;\n- default:\n- return false;\n- }\n+ return (_opcode == ParameterizedBuiltinFunctionOp.TRANSFORMENCODE);\n}\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/rewrite/RewriteMarkLoopVariablesUpdateInPlace.java", "new_path": "src/main/java/org/apache/sysml/hops/rewrite/RewriteMarkLoopVariablesUpdateInPlace.java", "diff": "@@ -99,12 +99,12 @@ public class RewriteMarkLoopVariablesUpdateInPlace extends StatementBlockRewrite\n//recursive invocation\nboolean ret = true;\nfor( StatementBlock sb : sbs ) {\n- if( !sb.variablesRead().containsVariable(varname) )\n+ if( !sb.variablesRead().containsVariable(varname)\n+ && !sb.variablesUpdated().containsVariable(varname) )\ncontinue; //valid wrt update-in-place\nif( sb instanceof WhileStatementBlock || sb instanceof ForStatementBlock ) {\n- ret &= sb.getUpdateInPlaceVars()\n- .contains(varname);\n+ ret &= sb.getUpdateInPlaceVars().contains(varname);\n}\nelse if( sb instanceof IfStatementBlock ) {\nIfStatementBlock isb = (IfStatementBlock) sb;\n" } ]

[ { "change_type": "MODIFY", "old_path": "bin/systemml-standalone.py", "new_path": "bin/systemml-standalone.py", "diff": "@@ -43,7 +43,7 @@ def default_classpath(systemml_home):\n#TODO\n# User dir, fix for SYSTEMML_1795\n-def standalone_execution_entry(nvargs, args, config, explain, debug, stats, gpu, f):\n+def standalone_execution_entry(nvargs, args, config, explain, debug, stats, gpu, heapmem, f):\n\"\"\"\nThis function is responsible for the execution of arguments via\nsubprocess call in singlenode mode\n@@ -57,7 +57,7 @@ def standalone_execution_entry(nvargs, args, config, explain, debug, stats, gpu,\nelse:\ndefault_cp = ':'.join(default_classpath(systemml_home))\n- java_memory = '-Xmx8g -Xms4g -Xmn1g'\n+ java_memory = '-Xmx' + heapmem + ' -Xms4g -Xmn1g'\n# Log4j\nlog4j = log4j_path(systemml_home)\n@@ -93,7 +93,10 @@ def standalone_execution_entry(nvargs, args, config, explain, debug, stats, gpu,\n'-f', script_file, '-exec', 'singlenode', '-config', default_config,\n' '.join(ml_options)]\n- return_code = os.system(' '.join(cmd))\n+ cmd = ' '.join(cmd)\n+ print(cmd)\n+\n+ return_code = os.system(cmd)\nreturn return_code\n@@ -115,6 +118,7 @@ if __name__ == '__main__':\ncparser.add_argument('-gpu', help='uses CUDA instructions when reasonable, '\n'set <force> option to skip conservative memory estimates '\n'and use GPU wherever possible', nargs='?')\n+ cparser.add_argument('-heapmem', help='maximum JVM heap memory', metavar='', default='8g')\ncparser.add_argument('-f', required=True, help='specifies dml/pydml file to execute; '\n'path can be local/hdfs/gpfs', metavar='')\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/run_perftest.py", "new_path": "scripts/perftest/python/run_perftest.py", "diff": "@@ -32,8 +32,7 @@ from datagen import config_packets_datagen\nfrom train import config_packets_train\nfrom predict import config_packets_predict\nfrom utils_misc import get_families, config_reader, \\\n- exec_dml_and_parse_time, exec_test_data, check_predict, get_folder_metrics, args_dict_split, \\\n- get_config_args\n+ exec_dml_and_parse_time, exec_test_data, check_predict, get_folder_metrics, split_config_args\nfrom utils_fs import create_dir_local, write_success, check_SUCCESS_file_exists\n# A packet is a dictionary\n@@ -84,8 +83,6 @@ ML_PREDICT = {'Kmeans': 'Kmeans-predict',\nDENSE_TYPE_ALGOS = ['clustering', 'stats1', 'stats2']\n-sup_args_dict = {}\n-\n# Responsible for execution and metric logging\ndef algorithm_workflow(algo, exec_type, config_path, dml_file_name, action_mode, current_dir):\n@@ -134,7 +131,7 @@ def algorithm_workflow(algo, exec_type, config_path, dml_file_name, action_mode,\nif exit_flag_success:\ntime = 'data_exists'\nelse:\n- time = exec_dml_and_parse_time(exec_type, dml_file_name, args, spark_args_dict, sup_args_dict, config_path)\n+ time = exec_dml_and_parse_time(exec_type, dml_file_name, args, backend_args_dict, systemml_args_dict, config_path)\nwrite_success(time, temp_cwd)\nprint('{},{},{},{},{},{}'.format(algo, action_mode, intercept, mat_type, mat_shape, time))\n@@ -222,7 +219,7 @@ def perf_test_entry(family, algo, exec_type, mat_type, mat_shape, config_dir, mo\n# Statistic family do not require to be split\nif family_name not in ['stats1', 'stats2']:\nif not success_file:\n- exec_test_data(exec_type, spark_args_dict, sup_args_dict, data_gen_dir, config)\n+ exec_test_data(exec_type, backend_args_dict, systemml_args_dict, data_gen_dir, config)\nif 'train' in mode:\n# Create config directories\n@@ -297,7 +294,7 @@ if __name__ == '__main__':\ndescription='SystemML Performance Test Script')\ncparser.add_argument('--family', help='space separated list of classes of algorithms '\n'(available : ' + ', '.join(sorted(all_families)) + ')',\n- metavar='', choices=all_families, nargs='+')\n+ metavar='', choices=all_families, nargs='+', default=' '.join(all_families))\ncparser.add_argument('--algo', help='space separated list of algorithm to run '\n'(Overrides --family, available : ' + ', '.join(sorted(all_algos)) + ')', metavar='',\nchoices=all_algos, nargs='+')\n@@ -329,7 +326,9 @@ if __name__ == '__main__':\ncparser.add_argument('-explain', help='explains plan levels can be hops, runtime, '\n'recompile_hops, recompile_runtime', nargs='?', const='runtime', metavar='')\ncparser.add_argument('-config', help='System-ML configuration file (e.g SystemML-config.xml)', metavar='')\n-\n+ cparser.add_argument('-gpu', help='uses CUDA instructions when reasonable, '\n+ 'set <force> option to skip conservative memory estimates '\n+ 'and use GPU wherever possible', nargs='?')\n# Spark Configuration Option\ncparser.add_argument('--master', help='local, yarn-client, yarn-cluster', metavar='')\ncparser.add_argument('--driver-memory', help='Memory for driver (e.g. 512M)', metavar='')\n@@ -338,15 +337,18 @@ if __name__ == '__main__':\ncparser.add_argument('--executor-cores', help='Number of cores', metavar='')\ncparser.add_argument('--conf', help='Spark configuration file', nargs='+', metavar='')\n+ # Single node execution mode options\n+ cparser.add_argument('-heapmem', help='maximum JVM heap memory', metavar='', default='8g')\n+\n+\n# Args is a namespace\nargs = cparser.parse_args()\nall_arg_dict = vars(args)\n- arg_dict, config_dict, spark_dict = args_dict_split(all_arg_dict)\ncreate_dir_local(args.config_dir)\n# Global variables\n- sup_args_dict, spark_args_dict = get_config_args(config_dict, spark_dict, args.exec_type)\n+ perftest_args_dict, systemml_args_dict, backend_args_dict = split_config_args(all_arg_dict)\n# Debug arguments\n# print(arg_dict)\n@@ -395,8 +397,8 @@ if __name__ == '__main__':\nlogging.info('algorithm,run_type,intercept,matrix_type,data_shape,time_sec')\n# Remove filename item from dictionary as its already used to create the log above\n- del arg_dict['filename']\n- perf_test_entry(**arg_dict)\n+ del perftest_args_dict['filename']\n+ perf_test_entry(**perftest_args_dict)\ntotal_time = (time.time() - start_time)\nlogging.info('Performance tests complete {0:.3f} secs \\n'.format(total_time))\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/utils_exec.py", "new_path": "scripts/perftest/python/utils_exec.py", "diff": "@@ -52,6 +52,7 @@ def subprocess_exec(cmd_string, log_file_path=None, extract=None):\nerror_arr, out_arr = get_all_logs(proc1)\nstd_outs = out_arr + error_arr\n+ std_outs.insert(0, ' '.join(exec_command))\nreturn_code = proc1.returncode\nif log_file_path is not None:\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/utils_misc.py", "new_path": "scripts/perftest/python/utils_misc.py", "diff": "@@ -30,53 +30,68 @@ from utils_exec import subprocess_exec\n# This file contains all misc utility functions required by performance test module\n-def get_config_args(config_dict, spark_dict, exec_type):\n+def split_config_args(args):\n\"\"\"\nBased on configuration parameters passed build configuration dictionary used by subprocess\n- config_dict: Dictionary\n- General configuration options\n+ args: Dictionary\n+ All parameters passed in\n- spark_dict: Dictionary\n- Spark configuration options\n+ return: Dictionary, Dictionary, Dictionary\n+ 3 dictionaries - one for perf tests, one for systemml specific args, one for backend options\n+ \"\"\"\n- exec_type: String\n- Contains the execution type singlenode / hybrid_spark\n+ perftest_args_dict = {}\n- return: Dictionary, Dictionary\n- Based on the parameters passed we build to dictionary that need to be passed either at the\n- beginning or at the end\n- \"\"\"\n+ perftest_args_dict['family'] = args['family']\n+ perftest_args_dict['algo'] = args['algo']\n+ perftest_args_dict['exec_type'] = args['exec_type']\n+ perftest_args_dict['mat_type'] = args['mat_type']\n+ perftest_args_dict['mat_shape'] = args['mat_shape']\n+ perftest_args_dict['config_dir'] = args['config_dir']\n+ perftest_args_dict['filename'] = args['filename']\n+ perftest_args_dict['mode'] = args['mode']\n+ perftest_args_dict['temp_dir'] = args['temp_dir']\n- sup_args_dict = {}\n- if config_dict['stats'] is not None:\n- sup_args_dict['-stats'] = config_dict['stats']\n- if config_dict['explain'] is not None:\n- sup_args_dict['-explain'] = config_dict['explain']\n+ systemml_args_dict = {}\n- if config_dict['config'] is not None:\n- sup_args_dict['-config'] = config_dict['config']\n+ if args['stats'] is not None:\n+ systemml_args_dict['-stats'] = args['stats']\n+\n+ if args['explain'] is not None:\n+ systemml_args_dict['-explain'] = args['explain']\n+\n+ if args['config'] is not None:\n+ systemml_args_dict['-config'] = args['config']\n+\n+ if args['gpu'] is not None:\n+ systemml_args_dict['-gpu'] = args['gpu']\n+\n+ backend_args_dict = {}\n+ exec_type = args['exec_type']\n- spark_args_dict = {}\nif exec_type == 'hybrid_spark':\n- if spark_dict['master'] is not None:\n- spark_args_dict['--master'] = spark_dict['master']\n+ if args['master'] is not None:\n+ backend_args_dict['--master'] = args['master']\n- if spark_dict['num_executors'] is not None:\n- spark_args_dict['--num-executors'] = spark_dict['num_executors']\n+ if args['num_executors'] is not None:\n+ backend_args_dict['--num-executors'] = args['num_executors']\n- if spark_dict['driver_memory'] is not None:\n- spark_args_dict['--driver-memory'] = spark_dict['driver_memory']\n+ if args['driver_memory'] is not None:\n+ backend_args_dict['--driver-memory'] = args['driver_memory']\n- if spark_dict['executor_cores'] is not None:\n- spark_args_dict['--executor-cores'] = spark_dict['executor_cores']\n+ if args['executor_cores'] is not None:\n+ backend_args_dict['--executor-cores'] = args['executor_cores']\n- if spark_dict['conf'] is not None:\n- spark_args_dict['--conf'] = ' '.join(spark_dict['conf'])\n+ if args['conf'] is not None:\n+ backend_args_dict['--conf'] = ' '.join(args['conf'])\n+ elif exec_type == 'singlenode':\n+ if args['heapmem'] is not None:\n+ backend_args_dict['-heapmem'] = args['heapmem']\n- return sup_args_dict, spark_args_dict\n+ return perftest_args_dict, systemml_args_dict, backend_args_dict\ndef args_dict_split(all_arguments):\n@@ -87,13 +102,14 @@ def args_dict_split(all_arguments):\nAll input arguments parsed\nreturn: Dictionary, Dictionary, Dictionary\n- We return three dictionaries for init, script, spark arguments\n+ We return four dictionaries for init, script, spark arguments, singlenode arguments\n\"\"\"\nargs_dict = dict(list(all_arguments.items())[0:9])\n- config_dict = dict(list(all_arguments.items())[9:12])\n- spark_dict = dict(list(all_arguments.items())[12:])\n+ config_dict = dict(list(all_arguments.items())[9:13])\n+ spark_dict = dict(list(all_arguments.items())[13:19])\n+ singlenode_dict = dict(list(all_arguments.items())[19:])\n- return args_dict, config_dict, spark_dict\n+ return args_dict, config_dict, spark_dict, singlenode_dict\ndef get_families(current_algo, ml_algo):\n@@ -166,7 +182,7 @@ def config_reader(read_path):\nreturn conf_file\n-def exec_dml_and_parse_time(exec_type, dml_file_name, args, spark_args_dict, sup_args_dict, log_file_name=None):\n+def exec_dml_and_parse_time(exec_type, dml_file_name, args, backend_args_dict, systemml_args_dict, log_file_name=None):\n\"\"\"\nThis function is responsible of execution of input arguments via python sub process,\nWe also extract time obtained from the output of this subprocess\n@@ -180,10 +196,10 @@ def exec_dml_and_parse_time(exec_type, dml_file_name, args, spark_args_dict, sup\nargs: Dictionary\nKey values pairs depending on the arg type\n- spark_args_dict: Dictionary\n- Spark configuration arguments\n+ backend_args_dict: Dictionary\n+ Spark configuration arguments / singlenode config arguments\n- sup_args_dict: Dictionary\n+ systemml_args_dict: Dictionary\nSupplementary arguments required by the script\nlog_file_name: String\n@@ -195,17 +211,17 @@ def exec_dml_and_parse_time(exec_type, dml_file_name, args, spark_args_dict, sup\nalgorithm = dml_file_name + '.dml'\n- sup_args = ''.join(['{} {}'.format(k, v) for k, v in sup_args_dict.items()])\n+ sup_args = ''.join(['{} {}'.format(k, v) for k, v in systemml_args_dict.items()])\nif exec_type == 'singlenode':\nexec_script = join(os.environ.get('SYSTEMML_HOME'), 'bin', 'systemml-standalone.py')\n-\n+ singlenode_pre_args = ''.join([' {} {} '.format(k, v) for k, v in backend_args_dict.items()])\nargs = ''.join(['{} {}'.format(k, v) for k, v in args.items()])\n- cmd = [exec_script, '-f', algorithm, args, sup_args]\n+ cmd = [exec_script, singlenode_pre_args, '-f', algorithm, args, sup_args]\ncmd_string = ' '.join(cmd)\nif exec_type == 'hybrid_spark':\nexec_script = join(os.environ.get('SYSTEMML_HOME'), 'bin', 'systemml-spark-submit.py')\n- spark_pre_args = ''.join([' {} {} '.format(k, v) for k, v in spark_args_dict.items()])\n+ spark_pre_args = ''.join([' {} {} '.format(k, v) for k, v in backend_args_dict.items()])\nargs = ''.join(['{} {}'.format(k, v) for k, v in args.items()])\ncmd = [exec_script, spark_pre_args, '-f', algorithm, args, sup_args]\ncmd_string = ' '.join(cmd)\n@@ -238,7 +254,7 @@ def parse_time(raw_logs):\nreturn 'time_not_found'\n-def exec_test_data(exec_type, spark_args_dict, sup_args_dict, datagen_path, config):\n+def exec_test_data(exec_type, backend_args_dict, systemml_args_dict, datagen_path, config):\n\"\"\"\nCreates the test data split from the given input path\n@@ -256,7 +272,7 @@ def exec_test_data(exec_type, spark_args_dict, sup_args_dict, datagen_path, conf\nX_test = join(path, 'X_test.data')\nY_test = join(path, 'Y_test.data')\nargs = {'-args': ' '.join([X, Y, X_test, Y_test, 'csv'])}\n- exec_dml_and_parse_time(exec_type, test_split_script, args, spark_args_dict, sup_args_dict)\n+ exec_dml_and_parse_time(exec_type, test_split_script, args, backend_args_dict, systemml_args_dict)\ndef check_predict(current_algo, ml_predict):\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/MANIFEST.in", "new_path": "src/main/python/MANIFEST.in", "diff": "#-------------------------------------------------------------\ninclude LICENSE\ninclude NOTICE\n+include README.txt\nrecursive-include systemml/systemml-java *\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/python/README.txt", "diff": "+Licensed to the Apache Software Foundation (ASF) under one or more\n+contributor license agreements. See the NOTICE file distributed with\n+this work for additional information regarding copyright ownership.\n+The ASF licenses this file to you under the Apache License, Version 2.0\n+(the \"License\"); you may not use this file except in compliance with\n+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, software\n+distributed under the License is distributed on an \"AS IS\" BASIS,\n+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n+See the License for the specific language governing permissions and\n+limitations under the License.\n+\n+\n+Apache SystemML\n+\n+Website: http://systemml.apache.org/\n+GitHub: https://github.com/apache/systemml\n+Mailing List: [email protected]\n+Issue Tracker: https://issues.apache.org/jira/browse/SYSTEMML\n+Download: http://systemml.apache.org/download.html\n+\n+For more information about using SystemML with Python, please see:\n+Beginner's Guide for Python Users: http://apache.github.io/systemml/beginners-guide-python.html\n+Reference Guide for Python Users: http://apache.github.io/systemml/python-reference.html\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/utils/GenerateClassesForMLContext.java", "new_path": "src/main/java/org/apache/sysml/utils/GenerateClassesForMLContext.java", "diff": "@@ -31,6 +31,9 @@ import java.util.Map;\nimport org.apache.commons.io.FileUtils;\nimport org.apache.commons.lang.StringEscapeUtils;\n+import org.apache.log4j.BasicConfigurator;\n+import org.apache.log4j.Level;\n+import org.apache.log4j.Logger;\nimport org.apache.sysml.api.DMLScript;\nimport org.apache.sysml.api.mlcontext.MLContext;\nimport org.apache.sysml.api.mlcontext.MLResults;\n@@ -89,6 +92,9 @@ public class GenerateClassesForMLContext {\nsource = args[0];\n}\ntry {\n+ BasicConfigurator.configure();\n+ Logger.getRootLogger().setLevel(Level.WARN);\n+\nDMLScript.VALIDATOR_IGNORE_ISSUES = true;\nSystem.out.println(\"************************************\");\nSystem.out.println(\"**** MLContext Class Generation ****\");\n" } ]

[ { "change_type": "MODIFY", "old_path": "bin/utils.py", "new_path": "bin/utils.py", "diff": "@@ -74,15 +74,16 @@ def find_dml_file(systemml_home, script_file):\nLocation of the dml script\n\"\"\"\nscripts_dir = join(systemml_home, 'scripts')\n- if not (exists(script_file)):\n- script_file = find_file(script_file, scripts_dir)\n- if script_file is None:\n+ if not exists(script_file):\n+ script_file_path = find_file(script_file, scripts_dir)\n+ if script_file_path is not None:\n+ return script_file_path\n+ else:\nprint('Could not find DML script: ' + script_file)\nsys.exit()\nreturn script_file\n-\ndef log4j_path(systemml_home):\n\"\"\"\nCreate log4j.properties from the template if not exist\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/run_perftest.py", "new_path": "scripts/perftest/python/run_perftest.py", "diff": "@@ -32,7 +32,8 @@ from datagen import config_packets_datagen\nfrom train import config_packets_train\nfrom predict import config_packets_predict\nfrom utils_misc import get_families, config_reader, \\\n- exec_dml_and_parse_time, exec_test_data, check_predict, get_folder_metrics, split_config_args\n+ exec_dml_and_parse_time, exec_test_data, check_predict, get_folder_metrics, split_config_args, \\\n+ get_default_dir\nfrom utils_fs import create_dir_local, write_success, check_SUCCESS_file_exists\n# A packet is a dictionary\n@@ -275,7 +276,7 @@ if __name__ == '__main__':\ndefault_mat_shape = ['10k_100']\n# Default temp directory, contains everything generated in perftest\n- default_temp_dir = join(systemml_home, 'scripts', 'perftest', 'temp')\n+ default_config_dir = join(systemml_home, 'scripts', 'perftest', 'temp')\n# Initialize time\nstart_time = time.time()\n@@ -308,7 +309,7 @@ if __name__ == '__main__':\ncparser.add_argument('--mat-shape', default=default_mat_shape, help='space separated list of shapes of matrices '\n'to generate (e.g 10k_1k, 20M_4k)', metavar='', nargs='+')\n- cparser.add_argument('--config-dir', default=default_temp_dir, help='temporary directory '\n+ cparser.add_argument('--config-dir', default=default_config_dir, help='temporary directory '\n'where generated, training and prediction data is put', metavar='')\ncparser.add_argument('--filename', default='perf_test', help='name of the output file for the perf'\n' metrics', metavar='')\n@@ -316,8 +317,7 @@ if __name__ == '__main__':\nhelp='space separated list of types of workloads to run (available: data-gen, train, predict)',\nmetavar='', choices=workload, nargs='+')\n# Change this to temp-dir\n- cparser.add_argument('--temp-dir', default=default_temp_dir,\n- help='define the file system to work on', metavar='')\n+ cparser.add_argument('--temp-dir', help='define the file system to work on', metavar='')\n# Configuration Options\ncparser.add_argument('-stats', help='Monitor and report caching/recompilation statistics, '\n@@ -350,8 +350,8 @@ if __name__ == '__main__':\n# Global variables\nperftest_args_dict, systemml_args_dict, backend_args_dict = split_config_args(all_arg_dict)\n- # Debug arguments\n- # print(arg_dict)\n+ # temp_dir hdfs / local path check\n+ perftest_args_dict['temp_dir'] = get_default_dir(args.temp_dir, args.exec_type, default_config_dir)\n# default_mat_type validity\nif len(args.mat_type) > 2:\n@@ -401,4 +401,5 @@ if __name__ == '__main__':\nperf_test_entry(**perftest_args_dict)\ntotal_time = (time.time() - start_time)\n- logging.info('Performance tests complete {0:.3f} secs \\n'.format(total_time))\n+ logging.info('total_time,none,none,none,none,{}'.format(total_time))\n+ logging.info('Performance tests complete')\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/utils_exec.py", "new_path": "scripts/perftest/python/utils_exec.py", "diff": "#\n#-------------------------------------------------------------\n+import sys\nimport subprocess\nimport shlex\nimport re\n@@ -63,6 +64,8 @@ def subprocess_exec(cmd_string, log_file_path=None, extract=None):\nreturn_data = parse_time(std_outs)\nif extract == 'dir':\nreturn_data = parse_hdfs_paths(std_outs)\n+ if extract == 'hdfs_base':\n+ return_data = parse_hdfs_base(std_outs)\nif extract is None:\nreturn_data = 0\n@@ -73,6 +76,20 @@ def subprocess_exec(cmd_string, log_file_path=None, extract=None):\nreturn return_data\n+def parse_hdfs_base(std_outs):\n+ \"\"\"\n+ return: String\n+ hdfs base uri\n+ \"\"\"\n+ hdfs_uri = None\n+ for line in std_outs:\n+ if line.startswith('hdfs://'):\n+ hdfs_uri = line\n+ if hdfs_uri is None:\n+ sys.exit('HDFS URI not found')\n+ return hdfs_uri\n+\n+\ndef write_logs(std_outs, log_file_path):\n\"\"\"\nWrite all logs to the specified location\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/utils_fs.py", "new_path": "scripts/perftest/python/utils_fs.py", "diff": "#-------------------------------------------------------------\nimport os\n+import sys\nfrom os.path import join\nimport glob\nfrom functools import reduce\n@@ -101,6 +102,16 @@ def contains_dir(hdfs_dirs, sub_folder):\nreturn False\n+def check_hdfs_path(path):\n+ \"\"\"\n+ Check if a path is present in HDFS\n+ \"\"\"\n+ cmd = ['hdfs', 'dfs', '-test', '-e', path]\n+ return_code = subprocess_exec(' '.join(cmd))\n+ if return_code != 0:\n+ return sys.exit('Please create {}'.format(path))\n+\n+\ndef relevant_folders(path, algo, family, matrix_type, matrix_shape, mode):\n\"\"\"\nFinds the right folder to read the data based on given parameters\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/utils_misc.py", "new_path": "scripts/perftest/python/utils_misc.py", "diff": "@@ -25,7 +25,9 @@ import os\nimport json\nimport re\nimport sys\n+import getpass\nfrom utils_exec import subprocess_exec\n+from utils_fs import check_hdfs_path\n# This file contains all misc utility functions required by performance test module\n@@ -361,3 +363,36 @@ def mat_type_check(current_family, matrix_types, dense_algos):\ncurrent_type.append(current_matrix_type)\nreturn current_type\n+\n+\n+def get_default_dir(temp_dir, exec_mode, config_dir):\n+ \"\"\"\n+ temp_dir: String\n+ exec_mode: String\n+ config_dir: String\n+\n+ return: String\n+ Local or HDFS home directory\n+ \"\"\"\n+\n+ if exec_mode == 'singlenode':\n+ if temp_dir is None:\n+ return config_dir\n+ if temp_dir is not None:\n+ return temp_dir\n+\n+ if exec_mode == 'hybrid_spark':\n+ cmd = ['hdfs', 'getconf', '-confKey', 'fs.default.name']\n+ hdfs_base = subprocess_exec(' '.join(cmd), extract='hdfs_base')\n+\n+ if temp_dir is None:\n+ hdfs_home = join(hdfs_base, 'user', getpass.getuser())\n+ check_hdfs_path(hdfs_home)\n+ return hdfs_home\n+\n+ if temp_dir is not None:\n+ if temp_dir.startswith('hdfs'):\n+ return temp_dir\n+ else:\n+ hdfs_home = join(hdfs_base, 'user', getpass.getuser(), temp_dir)\n+ return hdfs_home\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "</execution>\n</executions>\n</plugin>\n+ <!-- Attach python artifact so it can be installed and deployed. -->\n+ <plugin>\n+ <groupId>org.codehaus.mojo</groupId>\n+ <artifactId>build-helper-maven-plugin</artifactId>\n+ <version>1.8</version>\n+ <executions>\n+ <execution>\n+ <id>attach-python-artifact</id>\n+ <phase>pre-integration-test</phase>\n+ <goals>\n+ <goal>attach-artifact</goal>\n+ </goals>\n+ <configuration>\n+ <artifacts>\n+ <artifact>\n+ <file>${basedir}/target/${project.artifactId}-${project.version}-python.tgz</file>\n+ <type>tgz</type>\n+ <classifier>python</classifier>\n+ </artifact>\n+ </artifacts>\n+ </configuration>\n+ </execution>\n+ </executions>\n+ </plugin>\n</plugins>\n</build>\n</profile>\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/FrameBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/FrameBlock.java", "diff": "@@ -35,11 +35,11 @@ import java.util.Map;\nimport org.apache.commons.lang.ArrayUtils;\nimport org.apache.hadoop.io.Writable;\n-import org.apache.sysml.lops.Lop;\nimport org.apache.sysml.parser.Expression.ValueType;\nimport org.apache.sysml.runtime.DMLRuntimeException;\nimport org.apache.sysml.runtime.controlprogram.caching.CacheBlock;\nimport org.apache.sysml.runtime.io.IOUtilFunctions;\n+import org.apache.sysml.runtime.transform.encode.EncoderRecode;\nimport org.apache.sysml.runtime.util.IndexRange;\nimport org.apache.sysml.runtime.util.UtilFunctions;\n@@ -1049,12 +1049,7 @@ public class FrameBlock implements Writable, CacheBlock, Externalizable\nfor( int i=0; i<getNumRows(); i++ ) {\nObject val = ldata.get(i);\nif( val != null ) {\n- // Instead of using splitCSV which is forcing string with RFC-4180 format,\n- // using Lop.DATATYPE_PREFIX separator to split token and code\n- String[] tmp = new String[2];\n- int pos = val.toString().lastIndexOf(Lop.DATATYPE_PREFIX);\n- tmp[0] = val.toString().substring(0, pos);\n- tmp[1] = val.toString().substring(pos+1);\n+ String[] tmp = EncoderRecode.splitRecodeMapEntry(val.toString());\nmap.put(tmp[0], Long.parseLong(tmp[1]));\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/transform/decode/DecoderRecode.java", "new_path": "src/main/java/org/apache/sysml/runtime/transform/decode/DecoderRecode.java", "diff": "@@ -21,12 +21,12 @@ package org.apache.sysml.runtime.transform.decode;\nimport java.util.HashMap;\n-import org.apache.sysml.lops.Lop;\nimport org.apache.sysml.parser.Expression.ValueType;\nimport org.apache.sysml.runtime.matrix.data.FrameBlock;\nimport org.apache.sysml.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysml.runtime.matrix.data.Pair;\nimport org.apache.sysml.runtime.transform.TfUtils;\n+import org.apache.sysml.runtime.transform.encode.EncoderRecode;\nimport org.apache.sysml.runtime.util.UtilFunctions;\n/**\n@@ -82,7 +82,7 @@ public class DecoderRecode extends Decoder\nfor( int i=0; i<meta.getNumRows(); i++ ) {\nif( meta.get(i, _colList[j]-1)==null )\nbreak; //reached end of recode map\n- String[] tmp = meta.get(i, _colList[j]-1).toString().split(Lop.DATATYPE_PREFIX);\n+ String[] tmp = EncoderRecode.splitRecodeMapEntry(meta.get(i, _colList[j]-1).toString());\nObject obj = UtilFunctions.stringToObject(_schema[_colList[j]-1], tmp[0]);\nmap.put(Long.parseLong(tmp[1]), obj);\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/transform/encode/EncoderRecode.java", "new_path": "src/main/java/org/apache/sysml/runtime/transform/encode/EncoderRecode.java", "diff": "@@ -188,11 +188,25 @@ public class EncoderRecode extends Encoder\n/**\n* Returns the Recode map entry which consists of concatenation of code, delimiter and token.\n+ *\n* @param token is part of Recode map\n* @param code is code for token\n- * @return the concatenation of code and token with delimiter in between\n+ * @return the concatenation of token and code with delimiter in between\n*/\npublic static String constructRecodeMapEntry(String token, Long code) {\nreturn token + Lop.DATATYPE_PREFIX + code.toString();\n}\n+\n+ /**\n+ * Splits a Recode map entry into its token and code.\n+ *\n+ * @param value concatenation of token and code with delimiter in between\n+ * @return string array of token and code\n+ */\n+ public static String[] splitRecodeMapEntry(String value) {\n+ // Instead of using splitCSV which is forcing string with RFC-4180 format,\n+ // using Lop.DATATYPE_PREFIX separator to split token and code\n+ int pos = value.toString().lastIndexOf(Lop.DATATYPE_PREFIX);\n+ return new String[] {value.substring(0, pos), value.substring(pos+1)};\n+ }\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/ParameterizedBuiltinOp.java", "new_path": "src/main/java/org/apache/sysml/hops/ParameterizedBuiltinOp.java", "diff": "@@ -1025,15 +1025,15 @@ public class ParameterizedBuiltinOp extends Hop implements MultiThreadedHop\n//but very good sparsity estimate possible (number of non-zeros in input)\nHop max = getInput().get(_paramIndexMap.get(\"max\"));\nHop dir = getInput().get(_paramIndexMap.get(\"dir\"));\n- double maxVal = HopRewriteUtils.getDoubleValueSafe((LiteralOp)max);\n+ long maxVal = computeDimParameterInformation(max, memo);\nString dirVal = ((LiteralOp)dir).getStringValue();\nif( mc.dimsKnown() ) {\nlong lnnz = mc.nnzKnown() ? mc.getNonZeros() : mc.getRows();\nif( \"cols\".equals(dirVal) ) { //expand horizontally\n- ret = new long[]{mc.getRows(), UtilFunctions.toLong(maxVal), lnnz};\n+ ret = new long[]{mc.getRows(), maxVal, lnnz};\n}\nelse if( \"rows\".equals(dirVal) ){ //expand vertically\n- ret = new long[]{UtilFunctions.toLong(maxVal), mc.getRows(), lnnz};\n+ ret = new long[]{maxVal, mc.getRows(), lnnz};\n}\n}\n}\n@@ -1164,7 +1164,7 @@ public class ParameterizedBuiltinOp extends Hop implements MultiThreadedHop\nHop target = getInput().get(_paramIndexMap.get(\"target\"));\nHop max = getInput().get(_paramIndexMap.get(\"max\"));\nHop dir = getInput().get(_paramIndexMap.get(\"dir\"));\n- double maxVal = HopRewriteUtils.getDoubleValueSafe((LiteralOp)max);\n+ double maxVal = computeSizeInformation(max);\nString dirVal = ((LiteralOp)dir).getStringValue();\nif( \"cols\".equals(dirVal) ) { //expand horizontally\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/rewrite/HopRewriteUtils.java", "new_path": "src/main/java/org/apache/sysml/hops/rewrite/HopRewriteUtils.java", "diff": "@@ -456,6 +456,11 @@ public class HopRewriteUtils\nreturn datagen;\n}\n+ public static boolean isDataGenOp(Hop hop, DataGenMethod... ops) {\n+ return (hop instanceof DataGenOp\n+ && ArrayUtils.contains(ops, ((DataGenOp)hop).getOp()));\n+ }\n+\npublic static boolean isDataGenOpWithConstantValue(Hop hop, double value) {\nreturn hop instanceof DataGenOp\n&& ((DataGenOp)hop).getOp()==DataGenMethod.RAND\n@@ -989,17 +994,13 @@ public class HopRewriteUtils\nreturn ret;\n}\n- public static LiteralOp getBasic1NSequenceMaxLiteral(Hop hop)\n+ public static Hop getBasic1NSequenceMax(Hop hop)\nthrows HopsException\n{\n- if( hop instanceof DataGenOp )\n- {\n+ if( isDataGenOp(hop, DataGenMethod.SEQ) ) {\nDataGenOp dgop = (DataGenOp) hop;\n- if( dgop.getOp() == DataGenMethod.SEQ ){\n- Hop to = dgop.getInput().get(dgop.getParamIndex(Statement.SEQ_TO));\n- if( to instanceof LiteralOp )\n- return (LiteralOp)to;\n- }\n+ return dgop.getInput()\n+ .get(dgop.getParamIndex(Statement.SEQ_TO));\n}\nthrow new HopsException(\"Failed to retrieve 'to' argument from basic 1-N sequence.\");\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/rewrite/RewriteAlgebraicSimplificationStatic.java", "new_path": "src/main/java/org/apache/sysml/hops/rewrite/RewriteAlgebraicSimplificationStatic.java", "diff": "@@ -1641,7 +1641,7 @@ public class RewriteAlgebraicSimplificationStatic extends HopRewriteRule\n//setup input parameter hops\nHashMap<String,Hop> inputargs = new HashMap<String,Hop>();\ninputargs.put(\"target\", trgt);\n- inputargs.put(\"max\", HopRewriteUtils.getBasic1NSequenceMaxLiteral(seq));\n+ inputargs.put(\"max\", HopRewriteUtils.getBasic1NSequenceMax(seq));\ninputargs.put(\"dir\", new LiteralOp(direction));\ninputargs.put(\"ignore\", new LiteralOp(true));\ninputargs.put(\"cast\", new LiteralOp(false));\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/transform/encode/EncoderRecode.java", "new_path": "src/main/java/org/apache/sysml/runtime/transform/encode/EncoderRecode.java", "diff": "@@ -59,6 +59,8 @@ public class EncoderRecode extends Encoder\n}\nprivate long lookupRCDMap(int colID, String key) {\n+ if( !_rcdMaps.containsKey(colID) )\n+ return -1; //empty recode map\nLong tmp = _rcdMaps.get(colID).get(key);\nreturn (tmp!=null) ? tmp : -1;\n}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysml/test/integration/functions/transform/TransformApplyEmptyRecodeMapTest.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.sysml.test.integration.functions.transform;\n+\n+import org.junit.Assert;\n+import org.junit.Test;\n+import org.apache.sysml.parser.Expression.ValueType;\n+import org.apache.sysml.runtime.DMLRuntimeException;\n+import org.apache.sysml.runtime.matrix.data.FrameBlock;\n+import org.apache.sysml.runtime.matrix.data.MatrixBlock;\n+import org.apache.sysml.runtime.transform.encode.Encoder;\n+import org.apache.sysml.runtime.transform.encode.EncoderFactory;\n+import org.apache.sysml.runtime.util.DataConverter;\n+import org.apache.sysml.test.integration.AutomatedTestBase;\n+import org.apache.sysml.test.utils.TestUtils;\n+\n+public class TransformApplyEmptyRecodeMapTest extends AutomatedTestBase\n+{\n+ private static final int rows = 7;\n+ private static final int cols = 1;\n+\n+ @Override\n+ public void setUp() {\n+ TestUtils.clearAssertionInformation();\n+ }\n+\n+ @Test\n+ public void testTransformApplyEmptyRecodeMap() {\n+ try {\n+ //generate input data\n+ FrameBlock data = DataConverter.convertToFrameBlock(\n+ DataConverter.convertToMatrixBlock(getRandomMatrix(rows, cols, 1, 1, 1, 7)));\n+ FrameBlock meta = new FrameBlock(new ValueType[]{ValueType.STRING}, new String[]{\"C1\"});\n+\n+ //execute transform apply\n+ Encoder encoder = EncoderFactory.createEncoder(\n+ \"{ids:true, recode:[1]}\", data.getColumnNames(), meta.getSchema(), meta);\n+ MatrixBlock out = encoder.apply(data, new MatrixBlock(rows, cols, true));\n+\n+ //check outputs\n+ Assert.assertEquals(rows, out.getNumRows());\n+ Assert.assertEquals(cols, out.getNumColumns());\n+ for(int i=0; i<rows; i++)\n+ for(int j=0; j<cols; j++)\n+ Assert.assertTrue(Double.isNaN(out.quickGetValue(i, j)));\n+ }\n+ catch (DMLRuntimeException e) {\n+ throw new RuntimeException(e);\n+ }\n+ }\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/test_suites/java/org/apache/sysml/test/integration/functions/transform/ZPackageSuite.java", "new_path": "src/test_suites/java/org/apache/sysml/test/integration/functions/transform/ZPackageSuite.java", "diff": "@@ -27,6 +27,7 @@ import org.junit.runners.Suite;\n@RunWith(Suite.class)\[email protected]({\nFrameCSVReadWriteTest.class,\n+ TransformApplyEmptyRecodeMapTest.class,\nTransformCSVFrameEncodeDecodeTest.class,\nTransformCSVFrameEncodeReadTest.class,\nTransformEncodeDecodeTest.class,\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofCellwise.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofCellwise.java", "diff": "@@ -51,7 +51,6 @@ import org.apache.sysml.runtime.util.UtilFunctions;\npublic abstract class SpoofCellwise extends SpoofOperator implements Serializable\n{\nprivate static final long serialVersionUID = 3442528770573293590L;\n- private static final long PAR_NUMCELL_THRESHOLD = 1024*1024; //Min 1M elements\npublic enum CellType {\nNO_AGG,\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofMultiAggregate.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofMultiAggregate.java", "diff": "@@ -48,7 +48,6 @@ import org.apache.sysml.runtime.util.UtilFunctions;\npublic abstract class SpoofMultiAggregate extends SpoofOperator implements Serializable\n{\nprivate static final long serialVersionUID = -6164871955591089349L;\n- private static final long PAR_NUMCELL_THRESHOLD = 1024*1024; //Min 1M elements\nprivate final AggOp[] _aggOps;\nprivate final boolean _sparseSafe;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofOperator.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofOperator.java", "diff": "@@ -37,6 +37,10 @@ public abstract class SpoofOperator implements Serializable\nprivate static final long serialVersionUID = 3834006998853573319L;\nprivate static final Log LOG = LogFactory.getLog(SpoofOperator.class.getName());\n+ protected static final long PAR_NUMCELL_THRESHOLD = 1024*1024; //Min 1M elements\n+ protected static final long PAR_MINFLOP_THRESHOLD = 2L*1024*1024; //MIN 2 MFLOP\n+\n+\npublic abstract MatrixBlock execute(ArrayList<MatrixBlock> inputs, ArrayList<ScalarObject> scalars, MatrixBlock out)\nthrows DMLRuntimeException;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofOuterProduct.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofOuterProduct.java", "diff": "@@ -112,6 +112,9 @@ public abstract class SpoofOuterProduct extends SpoofOperator\nif( inputs.get(0).isEmptyBlock(false) )\nreturn new DoubleObject(0);\n+ if( 2*inputs.get(0).getNonZeros()*inputs.get(1).getNumColumns() < PAR_MINFLOP_THRESHOLD )\n+ return execute(inputs, scalarObjects); //sequential\n+\n//input preparation\ndouble[][] ab = getDenseMatrices(prepInputMatrices(inputs, 1, 2, true, false));\ndouble[][] b = getDenseMatrices(prepInputMatrices(inputs, 3, true));\n@@ -121,15 +124,14 @@ public abstract class SpoofOuterProduct extends SpoofOperator\nfinal int m = inputs.get(0).getNumRows();\nfinal int n = inputs.get(0).getNumColumns();\nfinal int k = inputs.get(1).getNumColumns(); // rank\n+ final long nnz = inputs.get(0).getNonZeros();\ndouble sum = 0;\ntry\n{\nExecutorService pool = Executors.newFixedThreadPool(k);\nArrayList<ParOuterProdAggTask> tasks = new ArrayList<ParOuterProdAggTask>();\n- //create tasks (for wdivmm-left, parallelization over columns;\n- //for wdivmm-right, parallelization over rows; both ensure disjoint results)\n- int numThreads2 = UtilFunctions.roundToNext(Math.min(8*k,m/32), k);\n+ int numThreads2 = getPreferredNumberOfTasks(m, n, nnz, k, numThreads);\nint blklen = (int)(Math.ceil((double)m/numThreads2));\nfor( int i=0; i<numThreads2 & i*blklen<m; i++ )\ntasks.add(new ParOuterProdAggTask(inputs.get(0), ab[0], ab[1], b, scalars,\n@@ -259,6 +261,9 @@ public abstract class SpoofOuterProduct extends SpoofOperator\nout.allocateDenseBlock();\n}\n+ if( 2*inputs.get(0).getNonZeros()*inputs.get(1).getNumColumns() < PAR_MINFLOP_THRESHOLD )\n+ return execute(inputs, scalarObjects, out); //sequential\n+\n//input preparation\ndouble[][] ab = getDenseMatrices(prepInputMatrices(inputs, 1, 2, true, false));\ndouble[][] b = getDenseMatrices(prepInputMatrices(inputs, 3, true));\n@@ -268,6 +273,7 @@ public abstract class SpoofOuterProduct extends SpoofOperator\nfinal int m = inputs.get(0).getNumRows();\nfinal int n = inputs.get(0).getNumColumns();\nfinal int k = inputs.get(1).getNumColumns(); // rank\n+ final long nnz = inputs.get(0).getNonZeros();\nMatrixBlock a = inputs.get(0);\n@@ -284,21 +290,24 @@ public abstract class SpoofOuterProduct extends SpoofOperator\nint numCG = ((CompressedMatrixBlock)a).getNumColGroups();\nint blklen = (int)(Math.ceil((double)numCG/numThreads));\nfor( int j=0; j<numThreads & j*blklen<numCG; j++ )\n- tasks.add(new ParExecTask(a, ab[0], ab[1], b, scalars, out, m, n, k, _outerProductType, 0, m, j*blklen, Math.min((j+1)*blklen, numCG)));\n+ tasks.add(new ParExecTask(a, ab[0], ab[1], b, scalars, out, m, n, k,\n+ _outerProductType, 0, m, j*blklen, Math.min((j+1)*blklen, numCG)));\n}\nelse {\n//parallelize over column partitions\nint blklen = (int)(Math.ceil((double)n/numThreads));\nfor( int j=0; j<numThreads & j*blklen<n; j++ )\n- tasks.add(new ParExecTask(a, ab[0], ab[1], b, scalars, out, m, n, k, _outerProductType, 0, m, j*blklen, Math.min((j+1)*blklen, n)));\n+ tasks.add(new ParExecTask(a, ab[0], ab[1], b, scalars, out, m, n, k,\n+ _outerProductType, 0, m, j*blklen, Math.min((j+1)*blklen, n)));\n}\n}\nelse { //right or cell-wise\n//parallelize over row partitions\n- int numThreads2 = UtilFunctions.roundToNext(Math.min(8*k,m/32), k);\n+ int numThreads2 = getPreferredNumberOfTasks(m, n, nnz, k, numThreads);\nint blklen = (int)(Math.ceil((double)m/numThreads2));\nfor( int i=0; i<numThreads2 & i*blklen<m; i++ )\n- tasks.add(new ParExecTask(a, ab[0], ab[1], b, scalars, out, m, n, k, _outerProductType, i*blklen, Math.min((i+1)*blklen,m), 0, n));\n+ tasks.add(new ParExecTask(a, ab[0], ab[1], b, scalars, out, m, n, k,\n+ _outerProductType, i*blklen, Math.min((i+1)*blklen,m), 0, n));\n}\nList<Future<Long>> taskret = pool.invokeAll(tasks);\npool.shutdown();\n@@ -320,6 +329,13 @@ public abstract class SpoofOuterProduct extends SpoofOperator\nreturn out;\n}\n+ private static int getPreferredNumberOfTasks(int m, int n, long nnz, int rank, int k) {\n+ //compute number of tasks nk in range [k, 8k]\n+ int base = (int) Math.min(Math.min(8*k, m/32),\n+ Math.ceil((double)2*nnz*rank/PAR_MINFLOP_THRESHOLD));\n+ return UtilFunctions.roundToNext(base, k);\n+ }\n+\nprivate void executeDense(double[] a, double[] u, double[] v, double[][] b, double[] scalars,\ndouble[] c, int m, int n, int k, OutProdType type, int rl, int ru, int cl, int cu )\n{\n@@ -427,6 +443,7 @@ public abstract class SpoofOuterProduct extends SpoofOperator\nif( !out.isInSparseFormat() ) //DENSE\n{\ndouble[] c = out.getDenseBlock();\n+ double tmp = 0;\nfor( int bi=rl; bi<ru; bi+=blocksizeIJ ) {\nint bimin = Math.min(ru, bi+blocksizeIJ);\n//prepare starting indexes for block row\n@@ -441,16 +458,20 @@ public abstract class SpoofOuterProduct extends SpoofOperator\nint[] wix = sblock.indexes(i);\ndouble[] wval = sblock.values(i);\nint index = wpos + curk[i-bi];\n- for( ; index<wpos+wlen && wix[index]<bjmin; index++ ) {\nif( type == OutProdType.CELLWISE_OUTER_PRODUCT )\n- c[wix[index]] = genexecCellwise( wval[index], u, uix, v, wix[index]*k, b, scalars, m, n, k, i, wix[index] );\n+ for( ; index<wpos+wlen && wix[index]<bjmin; index++ )\n+ c[wix[index]] = genexecCellwise( wval[index],\n+ u, uix, v, wix[index]*k, b, scalars, m, n, k, i, wix[index] );\nelse\n- c[0] += genexecCellwise( wval[index], u, uix, v, wix[index]*k, b, scalars, m, n, k, i, wix[index]);\n- }\n+ for( ; index<wpos+wlen && wix[index]<bjmin; index++ )\n+ tmp += genexecCellwise( wval[index],\n+ u, uix, v, wix[index]*k, b, scalars, m, n, k, i, wix[index]);\ncurk[i-bi] = index - wpos;\n}\n}\n}\n+ if( type != OutProdType.CELLWISE_OUTER_PRODUCT )\n+ c[0] = tmp;\n}\nelse //SPARSE\n{\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofRowwise.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofRowwise.java", "diff": "@@ -44,7 +44,6 @@ import org.apache.sysml.runtime.util.UtilFunctions;\npublic abstract class SpoofRowwise extends SpoofOperator\n{\nprivate static final long serialVersionUID = 6242910797139642998L;\n- private static final long PAR_NUMCELL_THRESHOLD = 1024*1024; //Min 1M elements\npublic enum RowType {\nNO_AGG, //no aggregation\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/BuiltinFunctionExpression.java", "new_path": "src/main/java/org/apache/sysml/parser/BuiltinFunctionExpression.java", "diff": "@@ -328,6 +328,9 @@ public class BuiltinFunctionExpression extends DataIdentifier\nDataIdentifier output = new DataIdentifier(outputName);\noutput.setAllPositions(this.getFilename(), this.getBeginLine(), this.getBeginColumn(), this.getEndLine(), this.getEndColumn());\n+ if (this.getFirstExpr() == null) {\n+ raiseValidateError(\"Function \" + this + \" has no arguments.\", false);\n+ }\nIdentifier id = this.getFirstExpr().getOutput();\noutput.setProperties(this.getFirstExpr().getOutput());\noutput.setNnz(-1); //conservatively, cannot use input nnz!\n@@ -1330,11 +1333,15 @@ public class BuiltinFunctionExpression extends DataIdentifier\n}\npublic String toString() {\n- StringBuilder sb = new StringBuilder(_opcode.toString() + \"(\" + _args[0].toString());\n- for(int i=1; i < _args.length; i++) {\n+ StringBuilder sb = new StringBuilder(_opcode.toString() + \"(\");\n+ if (_args != null) {\n+ for (int i = 0; i < _args.length; i++) {\n+ if (i > 0) {\nsb.append(\",\");\n+ }\nsb.append(_args[i].toString());\n}\n+ }\nsb.append(\")\");\nreturn sb.toString();\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/FunctionOp.java", "new_path": "src/main/java/org/apache/sysml/hops/FunctionOp.java", "diff": "@@ -167,6 +167,12 @@ public class FunctionOp extends Hop\nlong outputValues = OptimizerUtils.estimateSizeExactSparsity(getOutputs().get(1).getDim1(), 1, 1.0);\nreturn outputVectors+outputValues;\n}\n+ else if ( getFunctionName().equalsIgnoreCase(\"svd\") ) {\n+ long outputU = OptimizerUtils.estimateSizeExactSparsity(getOutputs().get(0).getDim1(), getOutputs().get(0).getDim2(), 1.0);\n+ long outputSigma = OptimizerUtils.estimateSizeExactSparsity(getOutputs().get(1).getDim1(), getOutputs().get(1).getDim2(), 1.0);\n+ long outputV = OptimizerUtils.estimateSizeExactSparsity(getOutputs().get(2).getDim1(), getOutputs().get(2).getDim2(), 1.0);\n+ return outputU+outputSigma+outputV;\n+ }\nelse\nthrow new RuntimeException(\"Invalid call of computeOutputMemEstimate in FunctionOp.\");\n}\n@@ -197,6 +203,10 @@ public class FunctionOp extends Hop\n//System.out.println(\"EigenInter \" + interOutput/1024/1024);\nreturn interOutput;\n}\n+ else if ( getFunctionName().equalsIgnoreCase(\"svd\")) {\n+ double interOutput = OptimizerUtils.estimateSizeExactSparsity(1, getInput().get(0).getDim2(), 1.0);\n+ return interOutput;\n+ }\nelse\nthrow new RuntimeException(\"Invalid call of computeIntermediateMemEstimate in FunctionOp.\");\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/Hop.java", "new_path": "src/main/java/org/apache/sysml/hops/Hop.java", "diff": "@@ -1024,6 +1024,7 @@ public abstract class Hop\nNOT, ABS, SIN, COS, TAN, ASIN, ACOS, ATAN, SIGN, SQRT, LOG, EXP,\nCAST_AS_SCALAR, CAST_AS_MATRIX, CAST_AS_FRAME, CAST_AS_DOUBLE, CAST_AS_INT, CAST_AS_BOOLEAN,\nPRINT, EIGEN, NROW, NCOL, LENGTH, ROUND, IQM, STOP, CEIL, FLOOR, MEDIAN, INVERSE, CHOLESKY,\n+ SVD,\n//cumulative sums, products, extreme values\nCUMSUM, CUMPROD, CUMMIN, CUMMAX,\n//fused ML-specific operators for performance\n@@ -1317,6 +1318,7 @@ public abstract class Hop\nHopsOpOp12String.put(OpOp1.CAST_AS_SCALAR, \"castAsScalar\");\nHopsOpOp12String.put(OpOp1.COS, \"cos\");\nHopsOpOp12String.put(OpOp1.EIGEN, \"eigen\");\n+ HopsOpOp12String.put(OpOp1.SVD, \"svd\");\nHopsOpOp12String.put(OpOp1.EXP, \"exp\");\nHopsOpOp12String.put(OpOp1.IQM, \"iqm\");\nHopsOpOp12String.put(OpOp1.MEDIAN, \"median\");\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/UnaryOp.java", "new_path": "src/main/java/org/apache/sysml/hops/UnaryOp.java", "diff": "@@ -693,7 +693,7 @@ public class UnaryOp extends Hop implements MultiThreadedHop\n//ensure cp exec type for single-node operations\nif( _op == OpOp1.PRINT || _op == OpOp1.STOP\n- || _op == OpOp1.INVERSE || _op == OpOp1.EIGEN || _op == OpOp1.CHOLESKY )\n+ || _op == OpOp1.INVERSE || _op == OpOp1.EIGEN || _op == OpOp1.CHOLESKY || _op == OpOp1.SVD)\n{\n_etype = ExecType.CP;\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/cost/CostEstimatorStaticRuntime.java", "new_path": "src/main/java/org/apache/sysml/hops/cost/CostEstimatorStaticRuntime.java", "diff": "@@ -994,13 +994,15 @@ public class CostEstimatorStaticRuntime extends CostEstimator\n//note: should be invoked independently for multiple outputs\nreturn d1m * d1n * d1s * DEFAULT_NFLOP_UNKNOWN;\n- case MultiReturnBuiltin: //opcodes: qr, lu, eigen\n+ case MultiReturnBuiltin: //opcodes: qr, lu, eigen, svd\n//note: they all have cubic complexity, the scaling factor refers to commons.math\ndouble xf = 2; //default e.g, qr\nif( optype.equals(\"eigen\") )\nxf = 32;\nelse if ( optype.equals(\"lu\") )\nxf = 16;\n+ else if ( optype.equals(\"svd\"))\n+ xf = 32; // TODO - assuming worst case for now\nreturn xf * d1m * d1n * d1n; //for 1kx1k ~ 2GFLOP -> 1s\ncase ParameterizedBuiltin: //opcodes: cdf, invcdf, groupedagg, rmempty\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/BuiltinFunctionExpression.java", "new_path": "src/main/java/org/apache/sysml/parser/BuiltinFunctionExpression.java", "diff": "@@ -200,6 +200,37 @@ public class BuiltinFunctionExpression extends DataIdentifier\nbreak;\n+ case SVD:\n+ checkNumParameters(1);\n+ checkMatrixParam(getFirstExpr());\n+\n+ long minMN = Math.min(getFirstExpr().getOutput().getDim1(), getFirstExpr().getOutput().getDim2());\n+\n+ // setup output properties\n+ DataIdentifier svdOut1 = (DataIdentifier) getOutputs()[0];\n+ DataIdentifier svdOut2 = (DataIdentifier) getOutputs()[1];\n+ DataIdentifier svdOut3 = (DataIdentifier) getOutputs()[2];\n+\n+ // Output 1\n+ svdOut1.setDataType(DataType.MATRIX);\n+ svdOut1.setValueType(ValueType.DOUBLE);\n+ svdOut1.setDimensions(getFirstExpr().getOutput().getDim1(), minMN);\n+ svdOut1.setBlockDimensions(getFirstExpr().getOutput().getRowsInBlock(), getFirstExpr().getOutput().getColumnsInBlock());\n+\n+ // Output 2\n+ svdOut2.setDataType(DataType.MATRIX);\n+ svdOut2.setValueType(ValueType.DOUBLE);\n+ svdOut2.setDimensions(minMN, minMN);\n+ svdOut2.setBlockDimensions(getFirstExpr().getOutput().getRowsInBlock(), getFirstExpr().getOutput().getColumnsInBlock());\n+\n+ // Output 3\n+ svdOut3.setDataType(DataType.MATRIX);\n+ svdOut3.setValueType(ValueType.DOUBLE);\n+ svdOut3.setDimensions(getFirstExpr().getOutput().getDim2(), minMN);\n+ svdOut3.setBlockDimensions(getFirstExpr().getOutput().getRowsInBlock(), getFirstExpr().getOutput().getColumnsInBlock());\n+\n+ break;\n+\ndefault: //always unconditional\nraiseValidateError(\"Unknown Builtin Function opcode: \" + _opcode, false);\n}\n@@ -1214,7 +1245,7 @@ public class BuiltinFunctionExpression extends DataIdentifier\nelse {\n// always unconditional (because unsupported operation)\nBuiltinFunctionOp op = getOpCode();\n- if( op==BuiltinFunctionOp.EIGEN || op==BuiltinFunctionOp.LU || op==BuiltinFunctionOp.QR )\n+ if( op==BuiltinFunctionOp.EIGEN || op==BuiltinFunctionOp.LU || op==BuiltinFunctionOp.QR || op==BuiltinFunctionOp.SVD)\nraiseValidateError(\"Function \"+op+\" needs to be called with multi-return assignment.\", false, LanguageErrorCodes.INVALID_PARAMETERS);\nelse\nraiseValidateError(\"Unsupported function \"+op, false, LanguageErrorCodes.INVALID_PARAMETERS);\n@@ -1256,6 +1287,7 @@ public class BuiltinFunctionExpression extends DataIdentifier\ncase QR:\ncase LU:\ncase EIGEN:\n+ case SVD:\nreturn true;\ndefault:\nreturn false;\n@@ -1667,6 +1699,8 @@ public class BuiltinFunctionExpression extends DataIdentifier\nbifop = Expression.BuiltinFunctionOp.INVERSE;\nelse if (functionName.equals(\"cholesky\"))\nbifop = Expression.BuiltinFunctionOp.CHOLESKY;\n+ else if (functionName.equals(\"svd\"))\n+ bifop = Expression.BuiltinFunctionOp.SVD;\nelse if (functionName.equals(\"sample\"))\nbifop = Expression.BuiltinFunctionOp.SAMPLE;\nelse if ( functionName.equals(\"outer\") )\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/DMLTranslator.java", "new_path": "src/main/java/org/apache/sysml/parser/DMLTranslator.java", "diff": "@@ -2471,6 +2471,7 @@ public class DMLTranslator\ncase QR:\ncase LU:\ncase EIGEN:\n+ case SVD:\n// Number of outputs = size of targetList = #of identifiers in source.getOutputs\nString[] outputNames = new String[targetList.size()];\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/Expression.java", "new_path": "src/main/java/org/apache/sysml/parser/Expression.java", "diff": "@@ -125,6 +125,7 @@ public abstract class Expression\nSOLVE,\nSQRT,\nSUM,\n+ SVD,\nTABLE,\nTAN,\nTRACE,\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/CPInstructionParser.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/CPInstructionParser.java", "diff": "@@ -267,11 +267,13 @@ public class CPInstructionParser extends InstructionParser\nString2CPInstructionType.put( \"qr\", CPINSTRUCTION_TYPE.MultiReturnBuiltin);\nString2CPInstructionType.put( \"lu\", CPINSTRUCTION_TYPE.MultiReturnBuiltin);\nString2CPInstructionType.put( \"eigen\", CPINSTRUCTION_TYPE.MultiReturnBuiltin);\n+ String2CPInstructionType.put( \"svd\", CPINSTRUCTION_TYPE.MultiReturnBuiltin);\nString2CPInstructionType.put( \"partition\", CPINSTRUCTION_TYPE.Partition);\nString2CPInstructionType.put( \"compress\", CPINSTRUCTION_TYPE.Compression);\nString2CPInstructionType.put( \"spoof\", CPINSTRUCTION_TYPE.SpoofFused);\n+\n//CP FILE instruction\nString2CPFileInstructionType = new HashMap<String, CPINSTRUCTION_TYPE>();\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/MultiReturnBuiltinCPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/MultiReturnBuiltinCPInstruction.java", "diff": "@@ -89,6 +89,17 @@ public class MultiReturnBuiltinCPInstruction extends ComputationCPInstruction\nreturn new MultiReturnBuiltinCPInstruction(null, in1, outputs, opcode, str);\n+ }\n+ else if ( opcode.equalsIgnoreCase(\"svd\") ) {\n+ CPOperand in1 = new CPOperand(parts[1]);\n+\n+ // one input and three outputs\n+ outputs.add ( new CPOperand(parts[2], ValueType.DOUBLE, DataType.MATRIX) );\n+ outputs.add ( new CPOperand(parts[3], ValueType.DOUBLE, DataType.MATRIX) );\n+ outputs.add ( new CPOperand(parts[4], ValueType.DOUBLE, DataType.MATRIX) );\n+\n+ return new MultiReturnBuiltinCPInstruction(null, in1, outputs, opcode, str);\n+\n}\nelse {\nthrow new DMLRuntimeException(\"Invalid opcode in MultiReturnBuiltin instruction: \" + opcode);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibCommonsMath.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibCommonsMath.java", "diff": "@@ -26,6 +26,7 @@ import org.apache.commons.math3.linear.EigenDecomposition;\nimport org.apache.commons.math3.linear.LUDecomposition;\nimport org.apache.commons.math3.linear.QRDecomposition;\nimport org.apache.commons.math3.linear.RealMatrix;\n+import org.apache.commons.math3.linear.SingularValueDecomposition;\nimport org.apache.sysml.runtime.DMLRuntimeException;\nimport org.apache.sysml.runtime.controlprogram.caching.MatrixObject;\nimport org.apache.sysml.runtime.util.DataConverter;\n@@ -48,7 +49,7 @@ public class LibCommonsMath\n}\npublic static boolean isSupportedMultiReturnOperation( String opcode ) {\n- return ( opcode.equals(\"qr\") || opcode.equals(\"lu\") || opcode.equals(\"eigen\") );\n+ return ( opcode.equals(\"qr\") || opcode.equals(\"lu\") || opcode.equals(\"eigen\") || opcode.equals(\"svd\") );\n}\npublic static boolean isSupportedMatrixMatrixOperation( String opcode ) {\n@@ -75,6 +76,8 @@ public class LibCommonsMath\nreturn computeLU(in);\nelse if (opcode.equals(\"eigen\"))\nreturn computeEigen(in);\n+ else if ( opcode.equals(\"svd\"))\n+ return computeSvd(in);\nreturn null;\n}\n@@ -217,6 +220,33 @@ public class LibCommonsMath\nreturn new MatrixBlock[] { mbValues, mbVectors };\n}\n+\n+ /**\n+ * Performs Singular Value Decomposition. Calls Apache Commons Math SVD.\n+ * X = U * Sigma * Vt, where X is the input matrix,\n+ * U is the left singular matrix, Sigma is the singular values matrix returned as a\n+ * column matrix and Vt is the transpose of the right singular matrix V.\n+ * However, the returned array has { U, Sigma, V}\n+ *\n+ * @param in Input matrix\n+ * @return An array containing U, Sigma & V\n+ * @throws DMLRuntimeException\n+ */\n+ private static MatrixBlock[] computeSvd(MatrixObject in) throws DMLRuntimeException {\n+ Array2DRowRealMatrix matrixInput = DataConverter.convertToArray2DRowRealMatrix(in);\n+\n+ SingularValueDecomposition svd = new SingularValueDecomposition(matrixInput);\n+ double[] sigma = svd.getSingularValues();\n+ RealMatrix u = svd.getU();\n+ RealMatrix v = svd.getV();\n+ MatrixBlock U = DataConverter.convertToMatrixBlock(u.getData());\n+ MatrixBlock Sigma = DataConverter.convertToMatrixBlock(sigma, true);\n+ Sigma = LibMatrixReorg.diag(Sigma, new MatrixBlock(Sigma.rlen, Sigma.rlen, true));\n+ MatrixBlock V = DataConverter.convertToMatrixBlock(v.getData());\n+\n+ return new MatrixBlock[] { U, Sigma, V };\n+ }\n+\n/**\n* Function to compute matrix inverse via matrix decomposition.\n*\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysml/test/integration/functions/unary/matrix/SVDFactorizeTest.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.sysml.test.integration.functions.unary.matrix;\n+\n+import org.junit.Test;\n+\n+import org.apache.sysml.api.DMLScript;\n+import org.apache.sysml.api.DMLScript.RUNTIME_PLATFORM;\n+import org.apache.sysml.runtime.matrix.MatrixCharacteristics;\n+import org.apache.sysml.test.integration.AutomatedTestBase;\n+import org.apache.sysml.test.integration.TestConfiguration;\n+\n+public class SVDFactorizeTest extends AutomatedTestBase\n+{\n+\n+ private final static String TEST_NAME1 = \"svd\";\n+ private final static String TEST_DIR = \"functions/unary/matrix/\";\n+ private static final String TEST_CLASS_DIR = TEST_DIR + SVDFactorizeTest.class.getSimpleName() + \"/\";\n+\n+ private final static int rows1 = 500;\n+ private final static int rows2 = 2500;\n+ private final static int cols1 = 400;\n+ private final static int cols2 = 2200;\n+ private final static double sparsity = 0.9;\n+\n+ @Override\n+ public void setUp()\n+ {\n+ addTestConfiguration(\n+ TEST_NAME1,\n+ new TestConfiguration(TEST_CLASS_DIR, TEST_NAME1,\n+ new String[] { \"D\" }) );\n+ }\n+\n+ @Test\n+ public void testSVDFactorizeDenseCP()\n+ {\n+ runTestSVDFactorize( rows1, cols1, RUNTIME_PLATFORM.SINGLE_NODE );\n+ }\n+\n+ @Test\n+ public void testSVDFactorizeDenseSP()\n+ {\n+ runTestSVDFactorize( rows1, cols1, RUNTIME_PLATFORM.SPARK );\n+ }\n+\n+ @Test\n+ public void testSVDFactorizeDenseMR()\n+ {\n+ runTestSVDFactorize( rows1, cols1, RUNTIME_PLATFORM.HADOOP );\n+ }\n+\n+ @Test\n+ public void testSVDFactorizeDenseHybrid()\n+ {\n+ runTestSVDFactorize( rows1, cols1, RUNTIME_PLATFORM.HYBRID );\n+ }\n+\n+ @Test\n+ public void testLargeSVDFactorizeDenseCP()\n+ {\n+ runTestSVDFactorize( rows2, cols2, RUNTIME_PLATFORM.SINGLE_NODE );\n+ }\n+\n+ @Test\n+ public void testLargeSVDFactorizeDenseSP()\n+ {\n+ runTestSVDFactorize( rows2, cols2, RUNTIME_PLATFORM.SPARK );\n+ }\n+\n+ @Test\n+ public void testLargeSVDFactorizeDenseMR()\n+ {\n+ runTestSVDFactorize( rows2, cols2, RUNTIME_PLATFORM.HADOOP );\n+ }\n+\n+ @Test\n+ public void testLargeSVDFactorizeDenseHybrid()\n+ {\n+ runTestSVDFactorize( rows2, cols2, RUNTIME_PLATFORM.HYBRID );\n+ }\n+\n+ private void runTestSVDFactorize( int rows, int cols, RUNTIME_PLATFORM rt)\n+ {\n+ RUNTIME_PLATFORM rtold = rtplatform;\n+ rtplatform = rt;\n+\n+ boolean sparkConfigOld = DMLScript.USE_LOCAL_SPARK_CONFIG;\n+ if( rtplatform == RUNTIME_PLATFORM.SPARK )\n+ DMLScript.USE_LOCAL_SPARK_CONFIG = true;\n+\n+ try\n+ {\n+ getAndLoadTestConfiguration(TEST_NAME1);\n+\n+ String HOME = SCRIPT_DIR + TEST_DIR;\n+ fullDMLScriptName = HOME + TEST_NAME1 + \".dml\";\n+ programArgs = new String[]{\"-args\", input(\"A\"), output(\"D\") };\n+\n+ double[][] A = getRandomMatrix(rows, cols, 0, 1, sparsity, 10);\n+ MatrixCharacteristics mc = new MatrixCharacteristics(rows, cols, -1, -1, -1);\n+ writeInputMatrixWithMTD(\"A\", A, false, mc);\n+\n+ // Expected matrix = 1x1 zero matrix\n+ double[][] D = new double[1][1];\n+ D[0][0] = 0.0;\n+ writeExpectedMatrix(\"D\", D);\n+\n+ boolean exceptionExpected = false;\n+ runTest(true, exceptionExpected, null, -1);\n+ compareResults(1e-8);\n+ }\n+ finally\n+ {\n+ rtplatform = rtold;\n+ }\n+ }\n+\n+}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/unary/matrix/svd.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+/*\n+ * DML script to test Singular Value Decomposition (SVD)\n+ */\n+\n+A = read($1);\n+\n+[U, S, V] = svd(A);\n+\n+# U(and V) must be orthogonal i.e., U = U^{-1} and V = V^{-1}\n+I1 = U %*% t(U);\n+I2 = V %*% t(V);\n+\n+Urows = nrow(U);\n+Ucols = ncol(U);\n+Vrows = nrow(V);\n+Vcols = ncol(V);\n+\n+# check for, whether the U and V are square or not.\n+#if ( Urows == Ucols & Vrows == Vcols) {\n+ a = sum(I1) / Urows;\n+ b = sum(I2) / Vrows;\n+#}\n+\n+D = matrix(1,1,1);\n+\n+if ( abs(1-a)<10e-4 & abs(1-b)<10e-4 ) {\n+\n+ # Multiplying U, S and t(V), we must get back original A\n+ B = U %*% S %*% t(V);\n+\n+ diff = sum(A - B);\n+ # write the difference between original A and computed B as a dummy 1x1 matrix\n+ D = diff*D;\n+\n+}\n+\n+write (D, $2);\n" }, { "change_type": "MODIFY", "old_path": "src/test_suites/java/org/apache/sysml/test/integration/functions/unary/matrix/ZPackageSuite.java", "new_path": "src/test_suites/java/org/apache/sysml/test/integration/functions/unary/matrix/ZPackageSuite.java", "diff": "@@ -42,6 +42,7 @@ import org.junit.runners.Suite;\nFullSignTest.class,\nIQMTest.class,\nLUFactorizeTest.class,\n+ SVDFactorizeTest.class,\nMatrixInverseTest.class,\nMinusTest.class,\nMLUnaryBuiltinTest.class,\n" } ]

[ { "change_type": "ADD", "old_path": null, "new_path": "samples/jupyter-notebooks/Image_Classify_Using_VGG_19_Transfer_Learning.ipynb", "diff": "+{\n+ \"cells\": [\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"# Image Classification using Caffe VGG-19 model (Transfer Learning)\\n\",\n+ \"\\n\",\n+ \"This notebook demonstrates importing VGG-19 model from Caffe to SystemML and use that model to do an image classification. VGG-19 model has been trained using ImageNet dataset (1000 classes with ~ 14M images). If an image to be predicted is in one of the class VGG-19 has trained on then accuracy will be higher.\\n\",\n+ \"We expect prediction of any image through SystemML using VGG-19 model will be similar to that of image predicted through Caffe using VGG-19 model directly.\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"#### Prerequisite:\\n\",\n+ \"1. SystemML Python Package\\n\",\n+ \"To run this notebook you need to install systeml 1.0 (Master Branch code as of 08/24/2017 or later) python package.\\n\",\n+ \"2. Download Dogs-vs-Cats Kaggle dataset from https://www.kaggle.com/c/dogs-vs-cats/data location to a directory.\\n\",\n+ \" Unzip the train.zip directory to some location and update the variable \\\"train_dir\\\" in bottom two cells in which classifyImagesWTransfLearning() and classifyImages() methods are called to test this change. \"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"##### SystemML Python Package information\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 1,\n+ \"metadata\": {\n+ \"collapsed\": false\n+ },\n+ \"outputs\": [\n+ {\n+ \"name\": \"stdout\",\n+ \"output_type\": \"stream\",\n+ \"text\": [\n+ \"Name: systemml\\r\\n\",\n+ \"Version: 1.0.0\\r\\n\",\n+ \"Summary: Apache SystemML is a distributed and declarative machine learning platform.\\r\\n\",\n+ \"Home-page: http://systemml.apache.org/\\r\\n\",\n+ \"Author: Apache SystemML\\r\\n\",\n+ \"Author-email: [email protected]\\r\\n\",\n+ \"License: Apache 2.0\\r\\n\",\n+ \"Location: /home/asurve/src/anaconda2/lib/python2.7/site-packages\\r\\n\",\n+ \"Requires: Pillow, numpy, scipy, pandas\\r\\n\"\n+ ]\n+ }\n+ ],\n+ \"source\": [\n+ \"!pip show systemml\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"#### SystemML Build information\\n\",\n+ \"Following code will show SystemML information which is installed in the environment.\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 2,\n+ \"metadata\": {\n+ \"collapsed\": false,\n+ \"scrolled\": true\n+ },\n+ \"outputs\": [\n+ {\n+ \"name\": \"stdout\",\n+ \"output_type\": \"stream\",\n+ \"text\": [\n+ \"SystemML Built-Time:2017-08-17 19:20:41 UTC\\n\",\n+ \"Archiver-Version: Plexus Archiver\\n\",\n+ \"Artifact-Id: systemml\\n\",\n+ \"Build-Jdk: 1.8.0_121\\n\",\n+ \"Build-Time: 2017-08-17 19:20:41 UTC\\n\",\n+ \"Built-By: asurve\\n\",\n+ \"Created-By: Apache Maven 3.3.9\\n\",\n+ \"Group-Id: org.apache.systemml\\n\",\n+ \"Main-Class: org.apache.sysml.api.DMLScript\\n\",\n+ \"Manifest-Version: 1.0\\n\",\n+ \"Minimum-Recommended-Spark-Version: 2.1.0\\n\",\n+ \"Version: 1.0.0-SNAPSHOT\\n\",\n+ \"\\n\"\n+ ]\n+ }\n+ ],\n+ \"source\": [\n+ \"from systemml import MLContext\\n\",\n+ \"ml = MLContext(sc)\\n\",\n+ \"print (\\\"SystemML Built-Time:\\\"+ ml.buildTime())\\n\",\n+ \"print(ml.info())\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 3,\n+ \"metadata\": {\n+ \"collapsed\": false,\n+ \"scrolled\": true\n+ },\n+ \"outputs\": [],\n+ \"source\": [\n+ \"# Workaround for Python 2.7.13 to avoid certificate validation issue while downloading any file.\\n\",\n+ \"\\n\",\n+ \"import ssl\\n\",\n+ \"\\n\",\n+ \"try:\\n\",\n+ \" _create_unverified_https_context = ssl._create_unverified_context\\n\",\n+ \"except AttributeError:\\n\",\n+ \" # Legacy Python that doesn't verify HTTPS certificates by default\\n\",\n+ \" pass\\n\",\n+ \"else:\\n\",\n+ \" # Handle target environment that doesn't support HTTPS verification\\n\",\n+ \" ssl._create_default_https_context = _create_unverified_https_context\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 4,\n+ \"metadata\": {\n+ \"collapsed\": true\n+ },\n+ \"outputs\": [],\n+ \"source\": [\n+ \"# Create label.txt file\\n\",\n+ \"\\n\",\n+ \"def createLabelFile(fileName):\\n\",\n+ \" file = open(fileName, 'w')\\n\",\n+ \" file.write('1,\\\"Cat\\\" \\\\n')\\n\",\n+ \" file.write('2,\\\"Dog\\\" \\\\n')\\n\",\n+ \" file.close()\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"#### Download model, proto files and convert them to SystemML format.\\n\",\n+ \"\\n\",\n+ \"1. Download Caffe Model (VGG-19), proto files (deployer, network and solver) and label file.\\n\",\n+ \"2. Convert the Caffe model into SystemML input format.\\n\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 5,\n+ \"metadata\": {\n+ \"collapsed\": false\n+ },\n+ \"outputs\": [],\n+ \"source\": [\n+ \"# Download caffemodel and proto files \\n\",\n+ \"\\n\",\n+ \"\\n\",\n+ \"def downloadAndConvertModel(downloadDir='.', trained_vgg_weights='trained_vgg_weights'):\\n\",\n+ \" \\n\",\n+ \" # Step 1: Download the VGG-19 model and other files.\\n\",\n+ \" import errno\\n\",\n+ \" import os\\n\",\n+ \" import urllib\\n\",\n+ \"\\n\",\n+ \" # Create directory, if exists don't error out\\n\",\n+ \" try:\\n\",\n+ \" os.makedirs(os.path.join(downloadDir,trained_vgg_weights))\\n\",\n+ \" except OSError as exc: # Python >2.5\\n\",\n+ \" if exc.errno == errno.EEXIST and os.path.isdir(trained_vgg_weights):\\n\",\n+ \" pass\\n\",\n+ \" else:\\n\",\n+ \" raise\\n\",\n+ \" \\n\",\n+ \" # Download deployer, network, solver proto and label files.\\n\",\n+ \" urllib.urlretrieve('https://raw.githubusercontent.com/apache/systemml/master/scripts/nn/examples/caffe2dml/models/imagenet/vgg19/VGG_ILSVRC_19_layers_deploy.proto', os.path.join(downloadDir,'VGG_ILSVRC_19_layers_deploy.proto'))\\n\",\n+ \" urllib.urlretrieve('https://raw.githubusercontent.com/apache/systemml/master/scripts/nn/examples/caffe2dml/models/imagenet/vgg19/VGG_ILSVRC_19_layers_network.proto',os.path.join(downloadDir,'VGG_ILSVRC_19_layers_network.proto'))\\n\",\n+ \" #TODO: After downloading network file (VGG_ILSVRC_19_layers_network.proto) , change num_output from 1000 to 2\\n\",\n+ \" \\n\",\n+ \" urllib.urlretrieve('https://raw.githubusercontent.com/apache/systemml/master/scripts/nn/examples/caffe2dml/models/imagenet/vgg19/VGG_ILSVRC_19_layers_solver.proto',os.path.join(downloadDir,'VGG_ILSVRC_19_layers_solver.proto'))\\n\",\n+ \" # TODO: set values as descrived below in VGG_ILSVRC_19_layers_solver.proto (Possibly through APIs whenever available)\\n\",\n+ \" # test_iter: 100\\n\",\n+ \" # stepsize: 40\\n\",\n+ \" # max_iter: 200\\n\",\n+ \" \\n\",\n+ \" # Create labels for data\\n\",\n+ \" ### 1,\\\"Cat\\\"\\n\",\n+ \" ### 2,\\\"Dog\\\"\\n\",\n+ \" createLabelFile(os.path.join(downloadDir, trained_vgg_weights, 'labels.txt'))\\n\",\n+ \"\\n\",\n+ \" # TODO: Following line commented as its 500MG file, if u need to download it please uncomment it and run.\\n\",\n+ \" # urllib.urlretrieve('http://www.robots.ox.ac.uk/~vgg/software/very_deep/caffe/VGG_ILSVRC_19_layers.caffemodel', os.path.join(downloadDir,'VGG_ILSVRC_19_layers.caffemodel'))\\n\",\n+ \"\\n\",\n+ \" # Step 2: Convert the caffemodel to trained_vgg_weights directory\\n\",\n+ \" import systemml as sml\\n\",\n+ \" sml.convert_caffemodel(sc, os.path.join(downloadDir,'VGG_ILSVRC_19_layers_deploy.proto'), os.path.join(downloadDir,'VGG_ILSVRC_19_layers.caffemodel'), os.path.join(downloadDir,trained_vgg_weights))\\n\",\n+ \" \\n\",\n+ \" return\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"##### PrintTopK\\n\",\n+ \"This function will print top K probabilities and indices from the result.\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 6,\n+ \"metadata\": {\n+ \"collapsed\": false\n+ },\n+ \"outputs\": [],\n+ \"source\": [\n+ \"# Print top K indices and probability\\n\",\n+ \"\\n\",\n+ \"def printTopK(prob, label, k):\\n\",\n+ \" print(label, 'Top ', k, ' Index : ', np.argsort(-prob)[0, :k])\\n\",\n+ \" print(label, 'Top ', k, ' Probability : ', prob[0,np.argsort(-prob)[0, :k]])\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"### Classify images\\n\",\n+ \"\\n\",\n+ \"This function classify images from images specified through urls.\\n\",\n+ \"\\n\",\n+ \"###### Input Parameters: \\n\",\n+ \" urls: List of urls\\n\",\n+ \" printTokKData (default False): Whether to print top K indices and probabilities\\n\",\n+ \" topK: Top K elements to be displayed. \"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 7,\n+ \"metadata\": {\n+ \"collapsed\": false\n+ },\n+ \"outputs\": [],\n+ \"source\": [\n+ \"import numpy as np\\n\",\n+ \"import urllib\\n\",\n+ \"from systemml.mllearn import Caffe2DML\\n\",\n+ \"import systemml as sml\\n\",\n+ \"\\n\",\n+ \"\\n\",\n+ \"def classifyImages(urls,img_shape=(3, 224, 224), printTokKData=False, topK=5, downloadDir='.', trained_vgg_weights='trained_vgg_weights'):\\n\",\n+ \"\\n\",\n+ \" size = (img_shape[1], img_shape[2])\\n\",\n+ \" \\n\",\n+ \" vgg = Caffe2DML(sqlCtx, solver=os.path.join(downloadDir,'VGG_ILSVRC_19_layers_solver.proto'), input_shape=img_shape)\\n\",\n+ \" vgg.load(trained_vgg_weights)\\n\",\n+ \"\\n\",\n+ \" for url in urls:\\n\",\n+ \" outFile = 'inputTest.jpg'\\n\",\n+ \" urllib.urlretrieve(url, outFile)\\n\",\n+ \" \\n\",\n+ \" from IPython.display import Image, display\\n\",\n+ \" display(Image(filename=outFile))\\n\",\n+ \" \\n\",\n+ \" print (\\\"Prediction of above image to ImageNet Class using\\\");\\n\",\n+ \"\\n\",\n+ \" ## Do image classification through SystemML processing\\n\",\n+ \" from PIL import Image\\n\",\n+ \" input_image = sml.convertImageToNumPyArr(Image.open(outFile), img_shape=img_shape\\n\",\n+ \" , color_mode='BGR', mean=sml.getDatasetMean('VGG_ILSVRC_19_2014'))\\n\",\n+ \" print (\\\"Image preprocessed through SystemML :: \\\", vgg.predict(input_image)[0])\\n\",\n+ \" if(printTopKData == True):\\n\",\n+ \" sysml_proba = vgg.predict_proba(input_image)\\n\",\n+ \" printTopK(sysml_proba, 'SystemML BGR', topK)\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 8,\n+ \"metadata\": {\n+ \"collapsed\": true\n+ },\n+ \"outputs\": [],\n+ \"source\": [\n+ \"from pyspark.ml.linalg import Vectors\\n\",\n+ \"import os\\n\",\n+ \"import systemml as sml\\n\",\n+ \"\\n\",\n+ \"\\n\",\n+ \"def getLabelFeatures(filename, train_dir, img_shape):\\n\",\n+ \" from PIL import Image\\n\",\n+ \"\\n\",\n+ \" vec = Vectors.dense(sml.convertImageToNumPyArr(Image.open(os.path.join(train_dir, filename)), img_shape=img_shape)[0,:])\\n\",\n+ \" if filename.lower().startswith('cat'):\\n\",\n+ \" return (1, vec)\\n\",\n+ \" elif filename.lower().startswith('dog'):\\n\",\n+ \" return (2, vec)\\n\",\n+ \" else:\\n\",\n+ \" raise ValueError('Expected the filename to start with either cat or dog')\\n\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 9,\n+ \"metadata\": {\n+ \"collapsed\": true\n+ },\n+ \"outputs\": [],\n+ \"source\": [\n+ \"from pyspark.sql.functions import rand\\n\",\n+ \"import os\\n\",\n+ \"\\n\",\n+ \"def createTrainingDF(train_dir, train_data_file, img_shape):\\n\",\n+ \" list_jpeg_files = os.listdir(train_dir)\\n\",\n+ \" # 10 files per partition\\n\",\n+ \" train_df = sc.parallelize(list_jpeg_files, int(len(list_jpeg_files)/10)).map(lambda filename : getLabelFeatures(filename, train_dir, img_shape)).toDF(['label', 'features']).orderBy(rand())\\n\",\n+ \" # Optional: but helps seperates conversion-related from training\\n\",\n+ \" # train_df.write.parquet(train_data_file) # 'kaggle-cats-dogs.parquet'\\n\",\n+ \" return train_df\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 10,\n+ \"metadata\": {\n+ \"collapsed\": true\n+ },\n+ \"outputs\": [],\n+ \"source\": [\n+ \"def readTrainingDF(train_dir, train_data_file):\\n\",\n+ \" train_df = sqlContext.read.parquet(train_data_file)\\n\",\n+ \" return train_df\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 11,\n+ \"metadata\": {\n+ \"collapsed\": true\n+ },\n+ \"outputs\": [],\n+ \"source\": [\n+ \"# downloadAndConvertModel(downloadDir, trained_vgg_weights)\\n\",\n+ \"# TODO: Take \\\"TODO\\\" actions mentioned in the downloadAndConvertModel() function after calling downloadAndConvertModel() function.\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 12,\n+ \"metadata\": {\n+ \"collapsed\": false\n+ },\n+ \"outputs\": [],\n+ \"source\": [\n+ \"def retrainModel(img_shape, downloadDir, trained_vgg_weights, train_dir, train_data_file, vgg_new_model):\\n\",\n+ \"\\n\",\n+ \" # Let downloadAndConvertModel() functon be commented out, as it needs to be called separately (which is done in cell above) and manual action to be taken after calling it.\\n\",\n+ \" # downloadAndConvertModel(downloadDir, trained_vgg_weights)\\n\",\n+ \" # TODO: Take \\\"TODO\\\" actions mentioned in the downloadAndConvertModel() function after calling that function.\\n\",\n+ \" \\n\",\n+ \" train_df = createTrainingDF(train_dir, train_data_file, img_shape)\\n\",\n+ \" ## Write from input files OR read if its already written/converted\\n\",\n+ \" # train_df = readTrainingDF(train_dir, train_data_file)\\n\",\n+ \" \\n\",\n+ \" # Load the model\\n\",\n+ \" vgg = Caffe2DML(sqlCtx, solver=os.path.join(downloadDir,'VGG_ILSVRC_19_layers_solver.proto'), input_shape=img_shape)\\n\",\n+ \" vgg.load(weights=os.path.join(downloadDir,trained_vgg_weights), ignore_weights=['fc8'])\\n\",\n+ \" vgg.set(debug=True).setExplain(True)\\n\",\n+ \"\\n\",\n+ \" # Train the model using new data\\n\",\n+ \" vgg.fit(train_df)\\n\",\n+ \" \\n\",\n+ \" # Save the trained model\\n\",\n+ \" vgg.save(vgg_new_model)\\n\",\n+ \" \\n\",\n+ \" return vgg\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": null,\n+ \"metadata\": {\n+ \"collapsed\": false\n+ },\n+ \"outputs\": [],\n+ \"source\": [\n+ \"import numpy as np\\n\",\n+ \"import urllib\\n\",\n+ \"from systemml.mllearn import Caffe2DML\\n\",\n+ \"import systemml as sml\\n\",\n+ \"\\n\",\n+ \"\\n\",\n+ \"def classifyImagesWTransfLearning(urls, model, img_shape=(3, 224, 224), printTokKData=False, topK=5):\\n\",\n+ \"\\n\",\n+ \" size = (img_shape[1], img_shape[2])\\n\",\n+ \" # vgg.load(trained_vgg_weights)\\n\",\n+ \"\\n\",\n+ \" for url in urls:\\n\",\n+ \" outFile = 'inputTest.jpg'\\n\",\n+ \" urllib.urlretrieve(url, outFile)\\n\",\n+ \" \\n\",\n+ \" from IPython.display import Image, display\\n\",\n+ \" display(Image(filename=outFile))\\n\",\n+ \" \\n\",\n+ \" print (\\\"Prediction of above image to ImageNet Class using\\\");\\n\",\n+ \"\\n\",\n+ \" ## Do image classification through SystemML processing\\n\",\n+ \" from PIL import Image\\n\",\n+ \" input_image = sml.convertImageToNumPyArr(Image.open(outFile), img_shape=img_shape\\n\",\n+ \" , color_mode='BGR', mean=sml.getDatasetMean('VGG_ILSVRC_19_2014'))\\n\",\n+ \"\\n\",\n+ \" print (\\\"Image preprocessed through SystemML :: \\\", model.predict(input_image)[0])\\n\",\n+ \" if(printTopKData == True):\\n\",\n+ \" sysml_proba = model.predict_proba(input_image)\\n\",\n+ \" printTopK(sysml_proba, 'SystemML BGR', topK)\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"### Sample code to retrain the model and use it to classify image through two different way\\n\",\n+ \"\\n\",\n+ \"There are couple of parameters to set based on what you are looking for.\\n\",\n+ \"1. printTopKData (default False): If this parameter gets set to True, then top K results (probabilities and indices) will be displayed. \\n\",\n+ \"2. topK (default 5): How many entities (K) to be displayed.\\n\",\n+ \"3. Directories, data file name, model name and directory where data has donwloaded.\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": null,\n+ \"metadata\": {\n+ \"collapsed\": false,\n+ \"scrolled\": false\n+ },\n+ \"outputs\": [],\n+ \"source\": [\n+ \"# ImageNet specific parameters\\n\",\n+ \"img_shape = (3, 224, 224)\\n\",\n+ \"\\n\",\n+ \"# Setting other than current directory causes \\\"network file not found\\\" issue, as network file\\n\",\n+ \"# location is defined in solver file which does not have a path, so it searches in current dir.\\n\",\n+ \"downloadDir = '.' # /home/asurve/caffe_models' \\n\",\n+ \"trained_vgg_weights = 'trained_vgg_weights'\\n\",\n+ \"\\n\",\n+ \"train_dir = '/home/asurve/data/keggle/dogs_vs_cats_2/train'\\n\",\n+ \"train_data_file = 'kaggle-cats-dogs.parquet'\\n\",\n+ \" \\n\",\n+ \"vgg_new_model = 'kaggle-cats-dogs-model_2'\\n\",\n+ \" \\n\",\n+ \"printTopKData=True\\n\",\n+ \"topK=5\\n\",\n+ \"\\n\",\n+ \"urls = ['http://cdn3-www.dogtime.com/assets/uploads/gallery/goldador-dog-breed-pictures/puppy-1.jpg','https://lh3.googleusercontent.com/-YdeAa1Ff4Ac/VkUnQ4vuZGI/AAAAAAAAAEg/nBiUn4pp6aE/w800-h800/images-6.jpeg','https://upload.wikimedia.org/wikipedia/commons/thumb/5/58/MountainLion.jpg/312px-MountainLion.jpg']\\n\",\n+ \"\\n\",\n+ \"vgg = retrainModel(img_shape, downloadDir, trained_vgg_weights, train_dir, train_data_file, vgg_new_model)\\n\",\n+ \"classifyImagesWTransfLearning(urls, vgg, img_shape, printTopKData, topK)\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": null,\n+ \"metadata\": {\n+ \"collapsed\": false,\n+ \"scrolled\": true\n+ },\n+ \"outputs\": [],\n+ \"source\": [\n+ \"img_shape = (3, 224, 224)\\n\",\n+ \"\\n\",\n+ \"printTopKData=True\\n\",\n+ \"topK=5\\n\",\n+ \"\\n\",\n+ \"# Setting other than current directory causes \\\"network file not found\\\" issue, as network file\\n\",\n+ \"# location is defined in solver file which does not have a path, so it searches in current dir.\\n\",\n+ \"downloadDir = '.' # /home/asurve/caffe_models' \\n\",\n+ \"trained_vgg_weights = 'kaggle-cats-dogs-model_2'\\n\",\n+ \"\\n\",\n+ \"urls = ['http://cdn3-www.dogtime.com/assets/uploads/gallery/goldador-dog-breed-pictures/puppy-1.jpg','https://lh3.googleusercontent.com/-YdeAa1Ff4Ac/VkUnQ4vuZGI/AAAAAAAAAEg/nBiUn4pp6aE/w800-h800/images-6.jpeg','https://upload.wikimedia.org/wikipedia/commons/thumb/5/58/MountainLion.jpg/312px-MountainLion.jpg']\\n\",\n+ \"\\n\",\n+ \"classifyImages(urls,img_shape, printTopKData, topK, downloadDir, trained_vgg_weights)\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": null,\n+ \"metadata\": {\n+ \"collapsed\": true\n+ },\n+ \"outputs\": [],\n+ \"source\": []\n+ }\n+ ],\n+ \"metadata\": {\n+ \"kernelspec\": {\n+ \"display_name\": \"Python 2\",\n+ \"language\": \"python\",\n+ \"name\": \"python2\"\n+ },\n+ \"language_info\": {\n+ \"codemirror_mode\": {\n+ \"name\": \"ipython\",\n+ \"version\": 2\n+ },\n+ \"file_extension\": \".py\",\n+ \"mimetype\": \"text/x-python\",\n+ \"name\": \"python\",\n+ \"nbconvert_exporter\": \"python\",\n+ \"pygments_lexer\": \"ipython2\",\n+ \"version\": \"2.7.13\"\n+ }\n+ },\n+ \"nbformat\": 4,\n+ \"nbformat_minor\": 2\n+}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/DMLProgram.java", "new_path": "src/main/java/org/apache/sysml/parser/DMLProgram.java", "diff": "@@ -33,8 +33,8 @@ public class DMLProgram\nprivate ArrayList<StatementBlock> _blocks;\nprivate HashMap<String, FunctionStatementBlock> _functionBlocks;\nprivate HashMap<String,DMLProgram> _namespaces;\n- public static String DEFAULT_NAMESPACE = \".defaultNS\";\n- public static String INTERNAL_NAMESPACE = \"_internal\"; // used for multi-return builtin functions\n+ public static final String DEFAULT_NAMESPACE = \".defaultNS\";\n+ public static final String INTERNAL_NAMESPACE = \"_internal\"; // used for multi-return builtin functions\nprivate static final Log LOG = LogFactory.getLog(DMLProgram.class.getName());\npublic DMLProgram(){\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/FunctionCallIdentifier.java", "new_path": "src/main/java/org/apache/sysml/parser/FunctionCallIdentifier.java", "diff": "@@ -168,7 +168,7 @@ public class FunctionCallIdentifier extends DataIdentifier\n@Override\npublic String toString() {\nStringBuilder sb = new StringBuilder();\n- if (_namespace != null && _namespace.length() > 0 && !_namespace.equals(\".defaultNS\"))\n+ if (_namespace != null && _namespace.length() > 0 && !_namespace.equals(DMLProgram.DEFAULT_NAMESPACE))\nsb.append(_namespace + \"::\");\nsb.append(_name);\nsb.append(\" ( \");\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/dml/DmlSyntacticValidator.java", "new_path": "src/main/java/org/apache/sysml/parser/dml/DmlSyntacticValidator.java", "diff": "@@ -852,13 +852,14 @@ public class DmlSyntacticValidator extends CommonSyntacticValidator implements D\nreturn;\n}\notherParams.put(paramName, val);\n- if(paramName.equals(\"classname\")) {\n+ if (paramName.equals(ExternalFunctionStatement.CLASS_NAME)) {\natleastOneClassName = true;\n}\n}\nfunctionStmt.setOtherParams(otherParams);\nif (!atleastOneClassName) {\n- notifyErrorListeners(\"the parameter \\'className\\' needs to be passed for externalFunction\", ctx.start);\n+ notifyErrorListeners(\"The \\'\" + ExternalFunctionStatement.CLASS_NAME\n+ + \"\\' argument needs to be passed to the externalFunction 'implemented in' clause.\", ctx.start);\nreturn;\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/assembly/bin.xml", "new_path": "src/assembly/bin.xml", "diff": "<dependencySet>\n<includes>\n<include>*:${artifactId}*</include>\n+ <include>*:avro*</include>\n<include>*:commons-cli*</include>\n<include>*:commons-collections*</include>\n<include>*:commons-configuration*</include>\n" }, { "change_type": "MODIFY", "old_path": "src/assembly/bin/LICENSE", "new_path": "src/assembly/bin/LICENSE", "diff": "The following components come under the Apache Software License 2.0.\n+avro-1.7.4.jar\n+avro-ipc-1.7.7-tests.jar\n+avro-ipc-1.7.7.jar\n+avro-mapred-1.7.7-hadoop2.jar\ncommons-cli-1.2.jar\ncommons-collections-3.2.1.jar\ncommons-configuration-1.6.jar\n" }, { "change_type": "MODIFY", "old_path": "src/assembly/standalone-jar.xml", "new_path": "src/assembly/standalone-jar.xml", "diff": "<dependencySet>\n<includes>\n<include>*:${artifactId}*</include>\n+ <include>*:avro*</include>\n<include>*:commons-cli*</include>\n<include>*:commons-collections*</include>\n<include>*:commons-configuration*</include>\n" }, { "change_type": "MODIFY", "old_path": "src/assembly/standalone-jar/LICENSE", "new_path": "src/assembly/standalone-jar/LICENSE", "diff": "@@ -214,6 +214,10 @@ commons-lang:commons-lang:2.6\ncommons-logging:commons-logging:1.1.3\nlog4j:log4j:1.2.15\nnet.sf.opencsv:opencsv:2.3\n+org.apache.avro:avro:1.7.4\n+org.apache.avro:avro-ipc:1.7.7\n+org.apache.avro:avro-ipc:1.7.7:tests\n+org.apache.avro:avro-mapred:1.7.7:hadoop2\norg.apache.commons:commons-math3:3.4.1\norg.apache.hadoop:hadoop-auth:2.6.0\norg.apache.hadoop:hadoop-client:2.6.0\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/StatementBlock.java", "new_path": "src/main/java/org/apache/sysml/parser/StatementBlock.java", "diff": "@@ -435,9 +435,9 @@ public class StatementBlock extends LiveVariableAnalysis implements ParseInfo\n}\nStatementBlock sblock = fstmt.getBody().get(0);\n- if( fcall.getParamExprs().size() < fstmt.getInputParams().size() ) {\n- sourceExpr.raiseValidateError(\"Wrong number of function parameters: \"+\n- fcall.getParamExprs().size() + \", but \" + fstmt.getInputParams().size()+\" expected.\");\n+ if( fcall.getParamExprs().size() != fstmt.getInputParams().size() ) {\n+ sourceExpr.raiseValidateError(\"Wrong number of function input arguments: \"+\n+ fcall.getParamExprs().size() + \" found, but \" + fstmt.getInputParams().size()+\" expected.\");\n}\nfor (int i =0; i < fstmt.getInputParams().size(); i++) {\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/ScriptExecutorUtils.java", "new_path": "src/main/java/org/apache/sysml/api/ScriptExecutorUtils.java", "diff": "@@ -80,6 +80,8 @@ public class ScriptExecutorUtils {\nDMLScript.FINEGRAINED_STATISTICS = dmlconf.getBooleanValue(DMLConfig.EXTRA_FINEGRAINED_STATS);\nDMLScript.STATISTICS_MAX_WRAP_LEN = dmlconf.getIntValue(DMLConfig.STATS_MAX_WRAP_LEN);\n+ boolean exceptionThrown = false;\n+\nStatistics.startRunTimer();\ntry {\n// run execute (w/ exception handling to ensure proper shutdown)\n@@ -93,6 +95,9 @@ public class ScriptExecutorUtils {\nec.setGPUContexts(gCtxs);\n}\nrtprog.execute(ec);\n+ } catch (Throwable e) {\n+ exceptionThrown = true;\n+ throw e;\n} finally { // ensure cleanup/shutdown\nif (DMLScript.USE_ACCELERATOR && !ec.getGPUContexts().isEmpty()) {\nec.getGPUContexts().forEach(gCtx -> gCtx.clearTemporaryMemory());\n@@ -104,10 +109,17 @@ public class ScriptExecutorUtils {\n// display statistics (incl caching stats if enabled)\nStatistics.stopRunTimer();\n+ if (!exceptionThrown) {\nif (statisticsMaxHeavyHitters > 0)\nSystem.out.println(Statistics.display(statisticsMaxHeavyHitters));\nelse\nSystem.out.println(Statistics.display());\n+ } else {\n+ if (statisticsMaxHeavyHitters > 0)\n+ System.err.println(Statistics.display(statisticsMaxHeavyHitters));\n+ else\n+ System.err.println(Statistics.display());\n+ }\n}\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/FunctionOp.java", "new_path": "src/main/java/org/apache/sysml/hops/FunctionOp.java", "diff": "@@ -22,6 +22,7 @@ package org.apache.sysml.hops;\nimport java.util.ArrayList;\nimport org.apache.sysml.lops.FunctionCallCP;\n+import org.apache.sysml.lops.FunctionCallCPSingle;\nimport org.apache.sysml.lops.Lop;\nimport org.apache.sysml.lops.LopsException;\nimport org.apache.sysml.lops.LopProperties.ExecType;\n@@ -33,14 +34,11 @@ import org.apache.sysml.runtime.controlprogram.parfor.opt.CostEstimatorHops;\n/**\n* This FunctionOp represents the call to a DML-bodied or external function.\n*\n- * Note: Currently, we support expressions in function arguments but no function calls\n- * in expressions.\n+ * Note: Currently, we support expressions in function arguments along with function calls\n+ * in expressions with single outputs, leaving multiple outputs handling as it is.\n*/\npublic class FunctionOp extends Hop\n{\n-\n- public static String OPSTRING = \"extfunct\";\n-\npublic enum FunctionType{\nDML,\nEXTERNAL_MEM,\n@@ -49,22 +47,25 @@ public class FunctionOp extends Hop\nUNKNOWN\n}\n+ public static final String OPSTRING = \"extfunct\";\n+\nprivate FunctionType _type = null;\nprivate String _fnamespace = null;\nprivate String _fname = null;\nprivate String[] _outputs = null;\nprivate ArrayList<Hop> _outputHops = null;\n+ private boolean _singleOutFun = false;\nprivate FunctionOp() {\n//default constructor for clone\n}\npublic FunctionOp(FunctionType type, String fnamespace, String fname, ArrayList<Hop> finputs, String[] outputs, ArrayList<Hop> outputHops) {\n- this(type, fnamespace, fname, finputs, outputs);\n+ this(type, fnamespace, fname, finputs, outputs, false);\n_outputHops = outputHops;\n}\n- public FunctionOp(FunctionType type, String fnamespace, String fname, ArrayList<Hop> finputs, String[] outputs)\n+ public FunctionOp(FunctionType type, String fnamespace, String fname, ArrayList<Hop> finputs, String[] outputs, boolean singleOut)\n{\nsuper(fnamespace + Program.KEY_DELIM + fname, DataType.UNKNOWN, ValueType.UNKNOWN );\n@@ -72,9 +73,9 @@ public class FunctionOp extends Hop\n_fnamespace = fnamespace;\n_fname = fname;\n_outputs = outputs;\n+ _singleOutFun = singleOut;\n- for( Hop in : finputs )\n- {\n+ for( Hop in : finputs ) {\ngetInput().add(in);\nin.getParent().add(this);\n}\n@@ -239,7 +240,8 @@ public class FunctionOp extends Hop\ntmp.add( in.constructLops() );\n//construct function call\n- FunctionCallCP fcall = new FunctionCallCP( tmp, _fnamespace, _fname, _outputs, _outputHops, et );\n+ Lop fcall = _singleOutFun ? new FunctionCallCPSingle( tmp, _fnamespace, _fname, et ) :\n+ new FunctionCallCP(tmp, _fnamespace, _fname, _outputs, _outputHops, et);\nsetLineNumbers(fcall);\nsetLops(fcall);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/lops/FunctionCallCP.java", "new_path": "src/main/java/org/apache/sysml/lops/FunctionCallCP.java", "diff": "@@ -22,6 +22,7 @@ package org.apache.sysml.lops;\nimport java.util.ArrayList;\n+import org.apache.sysml.hops.FunctionOp;\nimport org.apache.sysml.hops.Hop;\nimport org.apache.sysml.hops.HopsException;\nimport org.apache.sysml.lops.LopProperties.ExecLocation;\n@@ -122,7 +123,7 @@ public class FunctionCallCP extends Lop\ninst.append(getExecType());\ninst.append(Lop.OPERAND_DELIMITOR);\n- inst.append(\"extfunct\");\n+ inst.append(FunctionOp.OPSTRING);\ninst.append(Lop.OPERAND_DELIMITOR);\ninst.append(_fnamespace);\ninst.append(Lop.OPERAND_DELIMITOR);\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/main/java/org/apache/sysml/lops/FunctionCallCPSingle.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.sysml.lops;\n+\n+\n+import java.util.ArrayList;\n+\n+import org.apache.sysml.hops.FunctionOp;\n+import org.apache.sysml.lops.LopProperties.ExecLocation;\n+import org.apache.sysml.lops.LopProperties.ExecType;\n+import org.apache.sysml.lops.compile.JobType;\n+import org.apache.sysml.parser.DMLProgram;\n+import org.apache.sysml.parser.Expression.DataType;\n+import org.apache.sysml.parser.Expression.ValueType;\n+\n+public class FunctionCallCPSingle extends Lop\n+{\n+ private String _fnamespace;\n+ private String _fname;\n+\n+ public FunctionCallCPSingle(ArrayList<Lop> inputs, String fnamespace, String fname, ExecType et)\n+ {\n+ super(Lop.Type.FunctionCallCPSingle, DataType.UNKNOWN, ValueType.UNKNOWN);\n+ //note: data scalar in order to prevent generation of redundant createvar, rmvar\n+\n+ _fnamespace = fnamespace;\n+ _fname = fname;\n+\n+ //wire inputs\n+ for( Lop in : inputs ) {\n+ addInput( in );\n+ in.addOutput( this );\n+ }\n+\n+ //lop properties: always in CP\n+ boolean breaksAlignment = false;\n+ boolean aligner = false;\n+ boolean definesMRJob = false;\n+ lps.addCompatibility(JobType.INVALID);\n+ lps.setProperties(inputs, et, ExecLocation.ControlProgram, breaksAlignment, aligner, definesMRJob );\n+ }\n+\n+\n+ @Override\n+ public String toString() {\n+ return \"function call: \" + DMLProgram.constructFunctionKey(_fnamespace, _fname);\n+ }\n+\n+ @Override\n+ public String getInstructions(String input1, String output) throws LopsException {\n+ return getInstructions(new String[]{input1}, new String[]{output});\n+ }\n+\n+ @Override\n+ public String getInstructions(String input1, String input2, String output) throws LopsException {\n+ return getInstructions(new String[]{input1, input2}, new String[]{output});\n+ }\n+\n+ @Override\n+ public String getInstructions(String input1, String input2, String input3, String output) throws LopsException {\n+ return getInstructions(new String[]{input1, input2, input3}, new String[]{output});\n+ }\n+\n+ @Override\n+ public String getInstructions(String input1, String input2, String input3, String input4, String output) throws LopsException {\n+ return getInstructions(new String[]{input1, input2, input3, input4}, new String[]{output});\n+ }\n+\n+ @Override\n+ public String getInstructions(String input1, String input2, String input3, String input4, String input5, String output) throws LopsException {\n+ return getInstructions(new String[]{input1, input2, input3, input4, input5}, new String[]{output});\n+ }\n+\n+ @Override\n+ public String getInstructions(String input1, String input2, String input3, String input4, String input5, String input6, String output) throws LopsException {\n+ return getInstructions(new String[]{input1, input2, input3, input4, input5, input6}, new String[]{output});\n+ }\n+\n+ @Override\n+ public String getInstructions(String input1, String input2, String input3, String input4, String input5, String input6, String input7, String output) throws LopsException {\n+ return getInstructions(new String[]{input1, input2, input3, input4, input5, input6, input7}, new String[]{output});\n+ }\n+\n+ @Override\n+ public String getInstructions(String[] inputs, String output) throws LopsException\n+ {\n+ StringBuilder inst = new StringBuilder();\n+ inst.append(getExecType());\n+\n+ inst.append(Lop.OPERAND_DELIMITOR);\n+ inst.append(FunctionOp.OPSTRING);\n+ inst.append(Lop.OPERAND_DELIMITOR);\n+ inst.append(_fnamespace);\n+ inst.append(Lop.OPERAND_DELIMITOR);\n+ inst.append(_fname);\n+ inst.append(Lop.OPERAND_DELIMITOR);\n+ inst.append(inputs.length);\n+ inst.append(Lop.OPERAND_DELIMITOR);\n+ inst.append(\"1\"); //single output\n+\n+ for(int i=0; i<inputs.length; i++) {\n+ inst.append(Lop.OPERAND_DELIMITOR);\n+ inst.append( getInputs().get(i).prepInputOperand(inputs[i]) );\n+ }\n+\n+ inst.append(Lop.OPERAND_DELIMITOR);\n+ inst.append(output);\n+\n+ return inst.toString();\n+ }\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/lops/Lop.java", "new_path": "src/main/java/org/apache/sysml/lops/Lop.java", "diff": "@@ -52,7 +52,7 @@ public abstract class Lop\nCentralMoment, CoVariance, GroupedAgg, GroupedAggM,\nTransform, DataPartition, RepMat, //CP/MR reorganization, partitioning, replication\nParameterizedBuiltin, //CP/MR parameterized ops (name/value)\n- FunctionCallCP, //CP function calls\n+ FunctionCallCP, FunctionCallCPSingle, //CP function calls\nCumulativePartialAggregate, CumulativeSplitAggregate, CumulativeOffsetBinary, //MR cumsum/cumprod/cummin/cummax\nWeightedSquaredLoss, WeightedSigmoid, WeightedDivMM, WeightedCeMM, WeightedUMM,\nSortKeys, PickValues,\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/DMLTranslator.java", "new_path": "src/main/java/org/apache/sysml/parser/DMLTranslator.java", "diff": "@@ -1422,13 +1422,9 @@ public class DMLTranslator\n//create function op\nFunctionType ftype = fsb.getFunctionOpType();\n- FunctionOp fcall = null;\n- if (target == null) {\n- fcall = new FunctionOp(ftype, fci.getNamespace(), fci.getName(), finputs, new String[]{});\n- } else {\n- fcall = new FunctionOp(ftype, fci.getNamespace(), fci.getName(), finputs, new String[]{target.getName()});\n- }\n-\n+ FunctionOp fcall = (target == null) ?\n+ new FunctionOp(ftype, fci.getNamespace(), fci.getName(), finputs, new String[]{}, false) :\n+ new FunctionOp(ftype, fci.getNamespace(), fci.getName(), finputs, new String[]{target.getName()}, false);\noutput.add(fcall);\n//TODO function output dataops (phase 3)\n@@ -1465,7 +1461,7 @@ public class DMLTranslator\n}\nFunctionType ftype = fsb.getFunctionOpType();\n- FunctionOp fcall = new FunctionOp(ftype, fci.getNamespace(), fci.getName(), finputs, foutputs);\n+ FunctionOp fcall = new FunctionOp(ftype, fci.getNamespace(), fci.getName(), finputs, foutputs, false);\noutput.add(fcall);\n//TODO function output dataops (phase 3)\n" } ]

[ { "change_type": "MODIFY", "old_path": "docs/python-performance-test.md", "new_path": "docs/python-performance-test.md", "diff": "@@ -148,6 +148,17 @@ Run performance test for all algorithms under the family `regression2` and log w\nRun performance test for all algorithms using HDFS.\n+## Result Consolidation and Plotting\n+We have two scripts, `stats.py` forpulling results from google docs and `update.py` to updating results to google docs or local file system.\n+\n+Example of `update.py` would be below\n+`./scripts/perftest/python/google_docs/update.py --file ../../temp/perf_test_singlenode.out --exec-type singlenode --tag 2 --append test.csv`\n+The arguments being `--file` path of the perf-test output, `--exec-type` execution mode used to generate the perf-test output, `--tag` being the realease version or a unique name, `--append` being an optional argument that would append the a local csv file. If instead of `--append` the `--auth` argument needs the location of the `google api key` file.\n+\n+Example of `stats.py` below\n+` ./stats.py --auth ../key/client_json.json --exec-type singlenode --plot stats1_data-gen_none_dense_10k_100`\n+`--plot` argument needs the name of the composite key that you would like to compare results over. If this argument is not specified the results would be grouped by keys.\n+\n## Operational Notes\nAll performance test depend mainly on two scripts for execution `systemml-standalone.py` and `systemml-spark-submit.py`. Incase we need to change standalone or spark parameters we need to manually change these parameters in their respective scripts.\n@@ -158,7 +169,7 @@ The logs contain the following information below comma separated.\nalgorithm | run_type | intercept | matrix_type | data_shape | time_sec\n--- | --- | --- | --- | --- | --- |\n-multinomial|data-gen|0|dense|10k_100| 0.33\n+multinomial|data-gen|0|10k_100|dense| 0.33\nMultiLogReg|train|0|10k_100|dense|6.956\nMultiLogReg|predict|0|10k_100|dense|4.780\n@@ -187,9 +198,12 @@ Matrix Shape | Approximate Data Size\n10M_1k|80GB\n100M_1k|800GB\n+\nFor example the command below runs performance test for all data sizes described above\n`run_perftest.py --family binomial clustering multinomial regression1 regression2 stats1 stats2 --mat-shape 10k_1k 100k_1k 1M_1k 10M_1k 100M_1k --master yarn-client --temp-dir hdfs://localhost:9000/user/systemml`\n+By default data generated in `hybrid_spark` execution mode is in the current users `hdfs` home directory.\n+\nNote: Please use this command `pip3 install -r requirements.txt` before using the perftest scripts.\n" }, { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "<exclude>src/test/scripts/functions/jmlc/tfmtd_example/dummycoded.column.names</exclude>\n<exclude>src/test/scripts/functions/jmlc/tfmtd_example2/column.names</exclude>\n<exclude>src/test/scripts/functions/jmlc/tfmtd_frame_example/tfmtd_frame</exclude>\n+ <!-- Perftest requirement file -->\n+ <exclude>scripts/perftest/python/requirements.txt</exclude>\n</excludes>\n</configuration>\n</plugin>\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/datagen.py", "new_path": "scripts/perftest/python/datagen.py", "diff": "@@ -243,6 +243,7 @@ def config_packets_datagen(algo_payload, matrix_type, matrix_shape, datagen_dir,\nThis dictionary contains algorithms to be executed as keys and the path of configuration\njson files to be executed list of values.\n\"\"\"\n+\nconfig_bundle = {}\ndistinct_families = set(map(lambda x: x[1], algo_payload))\n" }, { "change_type": "ADD", "old_path": null, "new_path": "scripts/perftest/python/google_docs/gdocs_utils.py", "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+from oauth2client.service_account import ServiceAccountCredentials\n+import gspread\n+\n+\n+def auth(path, sheet_name):\n+ \"\"\"\n+ Responsible for authorization\n+ \"\"\"\n+\n+ scope = ['https://spreadsheets.google.com/feeds']\n+ creds = ServiceAccountCredentials.from_json_keyfile_name(path, scope)\n+ gc = gspread.authorize(creds)\n+ sheet = gc.open(\"Perf\").worksheet(sheet_name)\n+ return sheet\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/google_docs/stats.py", "new_path": "scripts/perftest/python/google_docs/stats.py", "diff": "# -------------------------------------------------------------\nimport argparse\n-from functools import reduce\n+import os\nimport pprint\n-from oauth2client.service_account import ServiceAccountCredentials\n-import gspread\n+from os.path import join\n+import matplotlib.pyplot as plt\n+from gdocs_utils import auth\n-# Get time difference between difference runs\n-\n-def auth(path, sheet_name):\n- \"\"\"\n- Responsible for authorization\n- \"\"\"\n- scope = ['https://spreadsheets.google.com/feeds']\n- creds = ServiceAccountCredentials.from_json_keyfile_name(path, scope)\n- gc = gspread.authorize(creds)\n- sheet = gc.open(\"Perf\").worksheet(sheet_name)\n- return sheet\n-\n-\n-def get_data(sheet, tag):\n+# Dict\n+# {algo_name : [algo_1.0': t1, 'algo_2.0': t2]}\n+def get_formatted_data(sheet_data):\n\"\"\"\n- Get time and algorithm from the sheet\n+ Read all the data from google sheets and transforms it into a dictionary that can be\n+ use for plotting later\n\"\"\"\n- time = sheet.find('time_{}'.format(tag))\n- algo = sheet.find('algo_{}'.format(tag))\n-\n- time_col = sheet.col_values(time.col)\n- time_col = list(filter(lambda x: len(x) > 0, time_col))\n-\n- algo_col = sheet.col_values(algo.col)\n- algo_col = list(filter(lambda x: len(x) > 0, algo_col))\n- return algo_col, time_col\n+ algo_dict = {}\n+\n+ for i in sheet_data:\n+ inn_count = 0\n+ data = []\n+ for key, val in i.items():\n+ inn_count += 1\n+ if inn_count < 3:\n+ data.append(key)\n+ data.append(val)\n+\n+ if inn_count == 2:\n+ t1, v1, _, v2 = data\n+ if len(str(v2)) > 0:\n+ if v1 not in algo_dict:\n+ algo_dict[v1] = [{t1: v2}]\n+ else:\n+ algo_dict[v1].append({t1: v2})\n+ inn_count = 0\n+ data = []\n+ return algo_dict\n-def get_data_dict(data_col):\n+def plot(x, y, xlab, ylab, title):\n\"\"\"\n- Return data as dictionary with key as algorithm and list time values\n+ Save plots to the current folder based on the arguments\n\"\"\"\n- data_dict = {}\n- all_algo = []\n- for algo, _ in data_col:\n- all_algo.append(algo)\n-\n- flatten_algo = reduce(lambda x, y: x+y, all_algo)\n-\n- # remove the header\n- filter_data = list(filter(lambda x: not x.startswith('algo_'), flatten_algo))\n- distict_algos = set(filter_data)\n-\n- for algo_dist in distict_algos:\n- for algo, time in data_col:\n- for k, v in zip(algo, time):\n- if algo_dist == k:\n- if algo_dist not in data_dict:\n- data_dict[k] = [v]\n- else:\n- data_dict[k].append(v)\n- return data_dict\n+ CWD = os.getcwd()\n+ PATH = join(CWD, title)\n+ width = .35\n+ plt.bar(x, y, color=\"red\", width=width)\n+ plt.xticks(x)\n+ plt.xlabel(xlab)\n+ plt.ylabel(ylab)\n+ plt.title(title)\n+ plt.savefig(PATH + '.png')\n+ print('Plot {} generated'.format(title))\n+ return plt\n# Example Usage\n-# ./stats.py --auth client_json.json --exec-mode singlenode --tags 1.0 2.0\n+# ./stats.py --auth ../key/client_json.json --exec-mode singlenode\nif __name__ == '__main__':\nexecution_mode = ['hybrid_spark', 'singlenode']\ncparser = argparse.ArgumentParser(description='System-ML Statistics Script')\ncparser.add_argument('--auth', help='Location to read auth file',\nrequired=True, metavar='')\n- cparser.add_argument('--exec-mode', help='Execution mode', choices=execution_mode,\n+ cparser.add_argument('--exec-type', help='Execution mode', choices=execution_mode,\nrequired=True, metavar='')\n- cparser.add_argument('--tags', help='Tagging header value',\n- required=True, nargs='+')\n+ cparser.add_argument('--plot', help='Algorithm to plot', metavar='')\nargs = cparser.parse_args()\n- arg_dict = vars(args)\n- sheet = auth(args.auth, args.exec_mode)\n- all_data = sheet.get_all_records()\n- data_col = []\n- for tag in args.tags:\n- algo_col, time_col = get_data(sheet, tag)\n- data_col.append((algo_col, time_col))\n-\n- data_dict = get_data_dict(data_col)\n-\n- delta_algo = {}\n- for k, v in data_dict.items():\n- delta = float(v[0]) - float(v[1])\n- delta_algo[k] = delta\n+ sheet = auth(args.auth, args.exec_type)\n+ all_data = sheet.get_all_records()\n- pprint.pprint(delta_algo, width=1)\n+ plot_data = get_formatted_data(all_data)\n+ if args.plot is not None:\n+ print(plot_data[args.plot])\n+ title = args.plot\n+ ylab = 'Time in sec'\n+ xlab = 'Version'\n+ x = []\n+ y = []\n+ for i in plot_data[args.plot]:\n+ version = list(i.keys())[0]\n+ time = list(i.values())[0]\n+ y.append(time)\n+ x.append(version)\n+\n+ x = list(map(lambda x: float(x.split('_')[1]), x))\n+ plot(x, y, xlab, ylab, title)\n+ else:\n+ pprint.pprint(plot_data, width=1)\n\\ No newline at end of file\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/google_docs/update.py", "new_path": "scripts/perftest/python/google_docs/update.py", "diff": "#\n# -------------------------------------------------------------\n+import sys\n+import os.path\nimport argparse\n-import gspread\n-from oauth2client.service_account import ServiceAccountCredentials\nimport pandas as pd\n-\n+from oauth2client.service_account import ServiceAccountCredentials\n+import gspread\n# Update data to google sheets\n@@ -33,6 +34,7 @@ def parse_data(file_path):\nSkip reading 1st row : Header\nSkip reading last row : Footer\n\"\"\"\n+\ncsv_file = pd.read_csv(file_path, sep=',', skiprows=1, skipfooter=1, engine='python')\nalgo = csv_file['INFO:root:algorithm'].apply(lambda x: x.split(':')[-1])\nkey = algo + '_'+ csv_file['run_type'] + '_' + csv_file['intercept'] + '_' + \\\n@@ -44,6 +46,7 @@ def auth(path, sheet_name):\n\"\"\"\nResponsible for authorization\n\"\"\"\n+\nscope = ['https://spreadsheets.google.com/feeds']\ncreds = ServiceAccountCredentials.from_json_keyfile_name(path, scope)\ngc = gspread.authorize(creds)\n@@ -64,6 +67,7 @@ def insert_values(sheet, key, col_num, header):\n\"\"\"\nInsert data to google sheets based on the arguments\n\"\"\"\n+\n# Col Name\nsheet.update_cell(1, col_num, header)\nfor id, val in enumerate(key):\n@@ -74,6 +78,7 @@ def get_dim(sheet):\n\"\"\"\nGet the dimensions of data\n\"\"\"\n+\ntry:\ncol_count = sheet.get_all_records()\nexcept:\n@@ -83,6 +88,16 @@ def get_dim(sheet):\nreturn row, col\n+def row_append(data_frame, file):\n+ \"\"\"\n+ Append results to a local csv\n+ \"\"\"\n+\n+ append_df = pd.read_csv(file)\n+ concat_data = pd.concat([data_frame, append_df], axis=1)\n+ return concat_data\n+\n+\n# Example Usage\n# ./update.py --file ../temp/test.out --exec-mode singlenode --auth client_json.json --tag 3.0\nif __name__ == '__main__':\n@@ -91,20 +106,36 @@ if __name__ == '__main__':\ncparser = argparse.ArgumentParser(description='System-ML Update / Stat Script')\ncparser.add_argument('--file', help='Location of the current perf test outputs',\nrequired=True, metavar='')\n- cparser.add_argument('--exec-mode', help='Backend Type', choices=execution_mode,\n- required=True, metavar='')\n- cparser.add_argument('--auth', help='Location to read auth file',\n+ cparser.add_argument('--exec-type', help='Backend Type', choices=execution_mode,\nrequired=True, metavar='')\ncparser.add_argument('--tag', help='Tagging header value',\nrequired=True, metavar='')\n+ cparser.add_argument('--auth', help='Location to read auth file', metavar='')\n+ cparser.add_argument('--append', help='Location to append the outputs', metavar='')\nargs = cparser.parse_args()\n- arg_dict = vars(args)\n- # Authenticate and get sheet dimensions\n- sheet = auth(args.auth, args.exec_mode)\n- row, col = get_dim(sheet)\n+ if args.auth is None and args.append is None:\n+ sys.exit('Both --auth and --append cannot be empty')\n+ algo, time = parse_data(args.file)\n+\n+ if args.append is not None:\n+ schema_df = {'algo_{}'.format(args.tag): algo,\n+ 'time_{}'.format(args.tag): time}\n+ data_frame = pd.DataFrame(schema_df)\n+ if os.path.isfile(args.append):\n+ append_data = row_append(data_frame, args.append)\n+ append_data.to_csv(args.append, sep=',', index=False)\n+ else:\n+ data_frame.to_csv(args.append, sep=',', index=False)\n+\n+ if args.auth is not None:\n# Read data from file and write to google docs\nalgo, time = parse_data(args.file)\n+\n+ # Authenticate and get sheet dimensions\n+ sheet = auth(args.auth, args.exec_type)\n+ row, col = get_dim(sheet)\n+\ninsert_pair(algo, time, col + 1, args.tag)\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/predict.py", "new_path": "scripts/perftest/python/predict.py", "diff": "@@ -239,10 +239,12 @@ def config_packets_predict(algo_payload, matrix_type, matrix_shape, datagen_dir,\nThis dictionary contains algorithms to be executed as keys and the path of configuration\njson files to be executed list of values.\n\"\"\"\n+\nconfig_bundle = {}\n- for k, _ in algo_payload:\n- config_bundle[k] = []\n+ for current_algo, current_family in algo_payload:\n+ key_name = current_algo + '.' + current_family\n+ config_bundle[key_name] = []\nfor current_algo, current_family in algo_payload:\ncurrent_matrix_type = mat_type_check(current_family, matrix_type, dense_algos)\n@@ -268,6 +270,7 @@ def config_packets_predict(algo_payload, matrix_type, matrix_shape, datagen_dir,\nconf_path = globals()[algo_func](save_name, current_data_gen_dir[0],\ncurrent_train_folder, predict_dir, config_dir)\n- config_bundle[current_algo].append(conf_path)\n+ key_name = current_algo + '.' + current_family\n+ config_bundle[key_name].append(conf_path)\nreturn config_bundle\n" }, { "change_type": "ADD", "old_path": null, "new_path": "scripts/perftest/python/requirements.txt", "diff": "+numpy\n+pandas\n+gspread\n+matplotlib\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/run_perftest.py", "new_path": "scripts/perftest/python/run_perftest.py", "diff": "@@ -233,10 +233,11 @@ def perf_test_entry(family, algo, exec_type, mat_type, mat_shape, config_dir, mo\nconf_packet = config_packets_train(algos_to_run, mat_type, mat_shape, data_gen_dir,\ntrain_dir, DENSE_TYPE_ALGOS, train_config_dir)\n- for algo_name, config_files in conf_packet.items():\n+ for algo_family_name, config_files in conf_packet.items():\nfor config in config_files:\n+ algo_name = algo_family_name.split('.')[0]\nfile_name = ML_TRAIN[algo_name]\n- algorithm_workflow(algo_name, exec_type, config, file_name, 'train', train_dir)\n+ algorithm_workflow(algo_family_name, exec_type, config, file_name, 'train', train_dir)\nif 'predict' in mode:\n# Create config directories\n@@ -255,10 +256,12 @@ def perf_test_entry(family, algo, exec_type, mat_type, mat_shape, config_dir, mo\nconf_packet = config_packets_predict(algos_to_run, mat_type, mat_shape, data_gen_dir,\ntrain_dir, predict_dir, DENSE_TYPE_ALGOS,\npredict_config_dir)\n- for algo_name, config_files in conf_packet.items():\n+\n+ for algo_family_name, config_files in conf_packet.items():\nfor config in config_files:\n+ algo_name = algo_family_name.split('.')[0]\nfile_name = ML_PREDICT[algo_name]\n- algorithm_workflow(algo_name, exec_type, config, file_name, 'predict', predict_dir)\n+ algorithm_workflow(algo_family_name, exec_type, config, file_name, 'predict', predict_dir)\nif __name__ == '__main__':\n@@ -290,12 +293,28 @@ if __name__ == '__main__':\n# Families\nall_families = ML_ALGO.keys()\n+ # Default Conf\n+ default_conf = 'spark.driver.maxResultSize=0 ' \\\n+ 'spark.akka.frameSize=128 ' \\\n+ 'spark.network.timeout=6000s ' \\\n+ 'spark.rpc.askTimeout=6000s ' \\\n+ 'spark.memory.useLegacyMode=true ' \\\n+ 'spark.files.useFetchCache=false' \\\n+\n+\n+ default_conf_big_job = 'spark.executor.extraJavaOptions=\\\"-Xmn5500m\\\" ' \\\n+ 'spark.executor.memory=\\\"-Xms50g\\\" ' \\\n+ 'spark.yarn.executor.memoryOverhead=8250 ' \\\n+ 'spark.driver.extraJavaOptions=\\\"-Xms20g -Xmn2g\\\"'\n+\n+\n+\n# Argparse Module\ncparser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter,\ndescription='SystemML Performance Test Script')\ncparser.add_argument('--family', help='space separated list of classes of algorithms '\n'(available : ' + ', '.join(sorted(all_families)) + ')',\n- metavar='', choices=all_families, nargs='+', default=' '.join(all_families))\n+ metavar='', choices=all_families, nargs='+', default=all_families)\ncparser.add_argument('--algo', help='space separated list of algorithm to run '\n'(Overrides --family, available : ' + ', '.join(sorted(all_algos)) + ')', metavar='',\nchoices=all_algos, nargs='+')\n@@ -335,12 +354,13 @@ if __name__ == '__main__':\ncparser.add_argument('--num-executors', help='Number of executors to launch', metavar='')\ncparser.add_argument('--executor-memory', help='Memory per executor', metavar='')\ncparser.add_argument('--executor-cores', help='Number of cores', metavar='')\n- cparser.add_argument('--conf', help='Spark configuration file', nargs='+', metavar='')\n+ cparser.add_argument('--conf', help='Spark configuration parameters, please use these '\n+ 'parameters for large performance tests ' + default_conf_big_job,\n+ default=default_conf, nargs='+', metavar='')\n# Single node execution mode options\ncparser.add_argument('-heapmem', help='maximum JVM heap memory', metavar='', default='8g')\n-\n# Args is a namespace\nargs = cparser.parse_args()\nall_arg_dict = vars(args)\n@@ -358,19 +378,7 @@ if __name__ == '__main__':\nprint('length of --mat-type argument cannot be greater than two')\nsys.exit()\n- # Check for validity of input arguments\n- if args.family is not None:\n- for fam in args.family:\n- if fam not in ML_ALGO.keys():\n- print('{} family not present in the performance test suit'.format(fam))\n- sys.exit()\n-\nif args.algo is not None:\n- for algo in args.algo:\n- if algo not in all_algos:\n- print('{} algorithm not present in the performance test suit'.format(args.algo))\n- sys.exit()\n-\n# This section check the validity of dual datagen algorithms like m-svm\nalgo_families = {}\nfor current_algo in args.algo:\n@@ -385,9 +393,8 @@ if __name__ == '__main__':\ninput_families = set(args.family)\ncommon_families = input_families.intersection(valid_families)\nif len(common_families) == 0:\n- print('Please specify a valid family for {} and the '\n+ sys.exit('Please specify a valid family for {} and the '\n'valid families are {}'.format(current_algo, ' '.join(valid_families)))\n- sys.exit()\n# Set level to 0 -> debug mode\n# Set level to 20 -> Plain metrics\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/train.py", "new_path": "scripts/perftest/python/train.py", "diff": "@@ -369,10 +369,12 @@ def config_packets_train(algo_payload, matrix_type, matrix_shape, datagen_dir, t\nThis dictionary contains algorithms to be executed as keys and the path of configuration\njson files to be executed list of values.\n\"\"\"\n+\nconfig_bundle = {}\n- for k, _ in algo_payload:\n- config_bundle[k] = []\n+ for current_algo, current_family in algo_payload:\n+ key_name = current_algo + '.' + current_family\n+ config_bundle[key_name] = []\nfor current_algo, current_family in algo_payload:\ncurrent_matrix_type = mat_type_check(current_family, matrix_type, dense_algos)\n@@ -389,7 +391,8 @@ def config_packets_train(algo_payload, matrix_type, matrix_shape, datagen_dir, t\nalgo_func = '_'.join([current_family] + [current_algo.lower().replace('-', '_')]\n+ ['train'])\nconf_path = globals()[algo_func](save_name, current_datagen_dir, train_dir, config_dir)\n- config_bundle[current_algo].append(conf_path)\n+ key_name = current_algo + '.' + current_family\n+ config_bundle[key_name].append(conf_path)\nconfig_packets = {}\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/utils_exec.py", "new_path": "scripts/perftest/python/utils_exec.py", "diff": "@@ -47,6 +47,7 @@ def subprocess_exec(cmd_string, log_file_path=None, extract=None):\n\"\"\"\n# Debug\n# print(cmd_string)\n+\nexec_command = shlex.split(cmd_string)\nproc1 = subprocess.Popen(exec_command, stdout=subprocess.PIPE,\nstderr=subprocess.PIPE)\n@@ -81,6 +82,7 @@ def parse_hdfs_base(std_outs):\nreturn: String\nhdfs base uri\n\"\"\"\n+\nhdfs_uri = None\nfor line in std_outs:\nif line.startswith('hdfs://'):\n@@ -94,6 +96,7 @@ def write_logs(std_outs, log_file_path):\n\"\"\"\nWrite all logs to the specified location\n\"\"\"\n+\nwith open(log_file_path, 'w')as log:\nlog.write(\"\\n\".join(std_outs))\n@@ -108,6 +111,7 @@ def get_all_logs(process):\nreturn: List, List\nStd out and Error as logs as list\n\"\"\"\n+\nout_arr = []\nwhile True:\nnextline = process.stdout.readline().decode('utf8').strip()\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/utils_fs.py", "new_path": "scripts/perftest/python/utils_fs.py", "diff": "@@ -37,6 +37,7 @@ def create_dir_local(directory):\ndirectory: String\nLocation to create a directory\n\"\"\"\n+\nif not os.path.exists(directory):\nos.makedirs(directory)\n@@ -51,6 +52,7 @@ def write_success(time, path):\npath: String\nLocation to write the SUCCESS file\n\"\"\"\n+\nif 'data-gen' in path:\nif path.startswith('hdfs') and len(time.split('.')) == 2:\nfull_path = join(path, '_SUCCESS')\n@@ -75,6 +77,7 @@ def check_SUCCESS_file_exists(path):\nreturn: Boolean\nChecks if the file _SUCCESS exists\n\"\"\"\n+\nif 'data-gen' in path:\nif path.startswith('hdfs'):\nfull_path = join(path, '_SUCCESS')\n@@ -93,6 +96,7 @@ def contains_dir(hdfs_dirs, sub_folder):\n\"\"\"\nSupport for Lambda Function to check if a HDFS subfolder is contained by the HDFS directory\n\"\"\"\n+\nif sub_folder in hdfs_dirs:\nreturn True\nelse:\n@@ -106,6 +110,7 @@ def check_hdfs_path(path):\n\"\"\"\nCheck if a path is present in HDFS\n\"\"\"\n+\ncmd = ['hdfs', 'dfs', '-test', '-e', path]\nreturn_code = subprocess_exec(' '.join(cmd))\nif return_code != 0:\n@@ -137,6 +142,7 @@ def relevant_folders(path, algo, family, matrix_type, matrix_shape, mode):\nreturn: List\nList of folder locations to read data from\n\"\"\"\n+\nfolders = []\nfor current_matrix_type in matrix_type:\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/utils_misc.py", "new_path": "scripts/perftest/python/utils_misc.py", "diff": "@@ -55,8 +55,6 @@ def split_config_args(args):\nperftest_args_dict['mode'] = args['mode']\nperftest_args_dict['temp_dir'] = args['temp_dir']\n-\n-\nsystemml_args_dict = {}\nif args['stats'] is not None:\n@@ -89,6 +87,7 @@ def split_config_args(args):\nif args['conf'] is not None:\nbackend_args_dict['--conf'] = ''.join(args['conf'])\n+\nelif exec_type == 'singlenode':\nif args['heapmem'] is not None:\nbackend_args_dict['-heapmem'] = args['heapmem']\n@@ -214,6 +213,7 @@ def exec_dml_and_parse_time(exec_type, dml_file_name, args, backend_args_dict, s\nalgorithm = dml_file_name + '.dml'\nsup_args = ''.join(['{} {}'.format(k, v) for k, v in systemml_args_dict.items()])\n+\nif exec_type == 'singlenode':\nexec_script = join(os.environ.get('SYSTEMML_HOME'), 'bin', 'systemml-standalone.py')\nsinglenode_pre_args = ''.join([' {} {} '.format(k, v) for k, v in backend_args_dict.items()])\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/QuaternaryOp.java", "new_path": "src/main/java/org/apache/sysml/hops/QuaternaryOp.java", "diff": "@@ -321,6 +321,74 @@ public class QuaternaryOp extends Hop implements MultiThreadedHop\nsetLops( wsloss );\n}\n+ private Lop obtainlU(Hop U, Hop V, boolean cacheU, double m1Size) throws HopsException, LopsException {\n+ Lop lU = null;\n+ if (cacheU) {\n+ // partitioning of U for read through distributed cache\n+ boolean needPartU = !U.dimsKnown() || U.getDim1() * U.getDim2() > DistributedCacheInput.PARTITION_SIZE;\n+ lU = U.constructLops();\n+ if (needPartU) { // requires partitioning\n+ lU = new DataPartition(lU, DataType.MATRIX, ValueType.DOUBLE,\n+ (m1Size > OptimizerUtils.getLocalMemBudget()) ? ExecType.MR : ExecType.CP,\n+ PDataPartitionFormat.ROW_BLOCK_WISE_N);\n+ lU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(), getRowsInBlock(), getColsInBlock(),\n+ U.getNnz());\n+ setLineNumbers(lU);\n+ }\n+ } else {\n+ // replication of U for shuffle to target block\n+ Lop offset = createOffsetLop(V, false); // ncol of t(V) -> nrow of V determines num replicates\n+ lU = new RepMat(U.constructLops(), offset, true, V.getDataType(), V.getValueType());\n+ lU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(), U.getRowsInBlock(), U.getColsInBlock(),\n+ U.getNnz());\n+ setLineNumbers(lU);\n+\n+ Group grpU = new Group(lU, Group.OperationTypes.Sort, DataType.MATRIX, ValueType.DOUBLE);\n+ grpU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(), U.getRowsInBlock(), U.getColsInBlock(),\n+ -1);\n+ setLineNumbers(grpU);\n+ lU = grpU;\n+ }\n+ return lU;\n+ }\n+\n+ private Lop obtainlV(Hop U, Hop V, boolean cacheV, double m2Size) throws HopsException, LopsException {\n+ Lop lV = null;\n+ if (cacheV) {\n+ // partitioning of V for read through distributed cache\n+ boolean needPartV = !V.dimsKnown() || V.getDim1() * V.getDim2() > DistributedCacheInput.PARTITION_SIZE;\n+ lV = V.constructLops();\n+ if (needPartV) { // requires partitioning\n+ lV = new DataPartition(lV, DataType.MATRIX, ValueType.DOUBLE,\n+ (m2Size > OptimizerUtils.getLocalMemBudget()) ? ExecType.MR : ExecType.CP,\n+ PDataPartitionFormat.ROW_BLOCK_WISE_N);\n+ lV.getOutputParameters().setDimensions(V.getDim1(), V.getDim2(), getRowsInBlock(), getColsInBlock(),\n+ V.getNnz());\n+ setLineNumbers(lV);\n+ }\n+ } else {\n+ // replication of t(V) for shuffle to target block\n+ Transform ltV = new Transform(V.constructLops(), HopsTransf2Lops.get(ReOrgOp.TRANSPOSE), getDataType(),\n+ getValueType(), ExecType.MR);\n+ ltV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(), V.getColsInBlock(), V.getRowsInBlock(),\n+ V.getNnz());\n+ setLineNumbers(ltV);\n+\n+ Lop offset = createOffsetLop(U, false); // nrow of U determines num replicates\n+ lV = new RepMat(ltV, offset, false, V.getDataType(), V.getValueType());\n+ lV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(), V.getColsInBlock(), V.getRowsInBlock(),\n+ V.getNnz());\n+ setLineNumbers(lV);\n+\n+ Group grpV = new Group(lV, Group.OperationTypes.Sort, DataType.MATRIX, ValueType.DOUBLE);\n+ grpV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(), V.getColsInBlock(), V.getRowsInBlock(),\n+ -1);\n+ setLineNumbers(grpV);\n+ lV = grpV;\n+ }\n+ return lV;\n+ }\n+\nprivate void constructMRLopsWeightedSquaredLoss(WeightsType wtype)\nthrows HopsException, LopsException\n{\n@@ -396,60 +464,8 @@ public class QuaternaryOp extends Hop implements MultiThreadedHop\nsetLineNumbers(grpW);\n}\n- Lop lU = null;\n- if( cacheU ) {\n- //partitioning of U for read through distributed cache\n- boolean needPartU = !U.dimsKnown() || U.getDim1() * U.getDim2() > DistributedCacheInput.PARTITION_SIZE;\n- lU = U.constructLops();\n- if( needPartU ){ //requires partitioning\n- lU = new DataPartition(lU, DataType.MATRIX, ValueType.DOUBLE, (m1Size>OptimizerUtils.getLocalMemBudget())?ExecType.MR:ExecType.CP, PDataPartitionFormat.ROW_BLOCK_WISE_N);\n- lU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(), getRowsInBlock(), getColsInBlock(), U.getNnz());\n- setLineNumbers(lU);\n- }\n- }\n- else {\n- //replication of U for shuffle to target block\n- Lop offset = createOffsetLop(V, false); //ncol of t(V) -> nrow of V determines num replicates\n- lU = new RepMat(U.constructLops(), offset, true, V.getDataType(), V.getValueType());\n- lU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(),\n- U.getRowsInBlock(), U.getColsInBlock(), U.getNnz());\n- setLineNumbers(lU);\n-\n- Group grpU = new Group(lU, Group.OperationTypes.Sort, DataType.MATRIX, ValueType.DOUBLE);\n- grpU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(), U.getRowsInBlock(), U.getColsInBlock(), -1);\n- setLineNumbers(grpU);\n- lU = grpU;\n- }\n-\n- Lop lV = null;\n- if( cacheV ) {\n- //partitioning of V for read through distributed cache\n- boolean needPartV = !V.dimsKnown() || V.getDim1() * V.getDim2() > DistributedCacheInput.PARTITION_SIZE;\n- lV = V.constructLops();\n- if( needPartV ){ //requires partitioning\n- lV = new DataPartition(lV, DataType.MATRIX, ValueType.DOUBLE, (m2Size>OptimizerUtils.getLocalMemBudget())?ExecType.MR:ExecType.CP, PDataPartitionFormat.ROW_BLOCK_WISE_N);\n- lV.getOutputParameters().setDimensions(V.getDim1(), V.getDim2(), getRowsInBlock(), getColsInBlock(), V.getNnz());\n- setLineNumbers(lV);\n- }\n- }\n- else {\n- //replication of t(V) for shuffle to target block\n- Transform ltV = new Transform( V.constructLops(), HopsTransf2Lops.get(ReOrgOp.TRANSPOSE), getDataType(), getValueType(), ExecType.MR);\n- ltV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(),\n- V.getColsInBlock(), V.getRowsInBlock(), V.getNnz());\n- setLineNumbers(ltV);\n-\n- Lop offset = createOffsetLop(U, false); //nrow of U determines num replicates\n- lV = new RepMat(ltV, offset, false, V.getDataType(), V.getValueType());\n- lV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(),\n- V.getColsInBlock(), V.getRowsInBlock(), V.getNnz());\n- setLineNumbers(lV);\n-\n- Group grpV = new Group(lV, Group.OperationTypes.Sort, DataType.MATRIX, ValueType.DOUBLE);\n- grpV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(), V.getColsInBlock(), V.getRowsInBlock(), -1);\n- setLineNumbers(grpV);\n- lV = grpV;\n- }\n+ Lop lU = obtainlU(U, V, cacheU, m1Size);\n+ Lop lV = obtainlV(U, V, cacheV, m2Size);\n//reduce-side wsloss w/ or without broadcast\nLop wsloss = new WeightedSquaredLossR(\n@@ -597,60 +613,8 @@ public class QuaternaryOp extends Hop implements MultiThreadedHop\ngrpX.getOutputParameters().setDimensions(X.getDim1(), X.getDim2(), X.getRowsInBlock(), X.getColsInBlock(), X.getNnz());\nsetLineNumbers(grpX);\n- Lop lU = null;\n- if( cacheU ) {\n- //partitioning of U for read through distributed cache\n- boolean needPartU = !U.dimsKnown() || U.getDim1() * U.getDim2() > DistributedCacheInput.PARTITION_SIZE;\n- lU = U.constructLops();\n- if( needPartU ){ //requires partitioning\n- lU = new DataPartition(lU, DataType.MATRIX, ValueType.DOUBLE, (m1Size>OptimizerUtils.getLocalMemBudget())?ExecType.MR:ExecType.CP, PDataPartitionFormat.ROW_BLOCK_WISE_N);\n- lU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(), getRowsInBlock(), getColsInBlock(), U.getNnz());\n- setLineNumbers(lU);\n- }\n- }\n- else {\n- //replication of U for shuffle to target block\n- Lop offset = createOffsetLop(V, false); //ncol of t(V) -> nrow of V determines num replicates\n- lU = new RepMat(U.constructLops(), offset, true, V.getDataType(), V.getValueType());\n- lU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(),\n- U.getRowsInBlock(), U.getColsInBlock(), U.getNnz());\n- setLineNumbers(lU);\n-\n- Group grpU = new Group(lU, Group.OperationTypes.Sort, DataType.MATRIX, ValueType.DOUBLE);\n- grpU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(), U.getRowsInBlock(), U.getColsInBlock(), -1);\n- setLineNumbers(grpU);\n- lU = grpU;\n- }\n-\n- Lop lV = null;\n- if( cacheV ) {\n- //partitioning of V for read through distributed cache\n- boolean needPartV = !V.dimsKnown() || V.getDim1() * V.getDim2() > DistributedCacheInput.PARTITION_SIZE;\n- lV = V.constructLops();\n- if( needPartV ){ //requires partitioning\n- lV = new DataPartition(lV, DataType.MATRIX, ValueType.DOUBLE, (m2Size>OptimizerUtils.getLocalMemBudget())?ExecType.MR:ExecType.CP, PDataPartitionFormat.ROW_BLOCK_WISE_N);\n- lV.getOutputParameters().setDimensions(V.getDim1(), V.getDim2(), getRowsInBlock(), getColsInBlock(), V.getNnz());\n- setLineNumbers(lV);\n- }\n- }\n- else {\n- //replication of t(V) for shuffle to target block\n- Transform ltV = new Transform( V.constructLops(), HopsTransf2Lops.get(ReOrgOp.TRANSPOSE), getDataType(), getValueType(), ExecType.MR);\n- ltV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(),\n- V.getColsInBlock(), V.getRowsInBlock(), V.getNnz());\n- setLineNumbers(ltV);\n-\n- Lop offset = createOffsetLop(U, false); //nrow of U determines num replicates\n- lV = new RepMat(ltV, offset, false, V.getDataType(), V.getValueType());\n- lV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(),\n- V.getColsInBlock(), V.getRowsInBlock(), V.getNnz());\n- setLineNumbers(lV);\n-\n- Group grpV = new Group(lV, Group.OperationTypes.Sort, DataType.MATRIX, ValueType.DOUBLE);\n- grpV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(), V.getColsInBlock(), V.getRowsInBlock(), -1);\n- setLineNumbers(grpV);\n- lV = grpV;\n- }\n+ Lop lU = obtainlU(U, V, cacheU, m1Size);\n+ Lop lV = obtainlV(U, V, cacheV, m2Size);\n//reduce-side wsig w/ or without broadcast\nLop wsigmoid = new WeightedSigmoidR(\n@@ -793,60 +757,8 @@ public class QuaternaryOp extends Hop implements MultiThreadedHop\ngrpX.getOutputParameters().setDimensions(X.getDim1(), X.getDim2(), X.getRowsInBlock(), X.getColsInBlock(), X.getNnz());\nsetLineNumbers(grpX);\n- Lop lU = null;\n- if( cacheU ) {\n- //partitioning of U for read through distributed cache\n- boolean needPartU = !U.dimsKnown() || U.getDim1() * U.getDim2() > DistributedCacheInput.PARTITION_SIZE;\n- lU = U.constructLops();\n- if( needPartU ){ //requires partitioning\n- lU = new DataPartition(lU, DataType.MATRIX, ValueType.DOUBLE, (m1Size>OptimizerUtils.getLocalMemBudget())?ExecType.MR:ExecType.CP, PDataPartitionFormat.ROW_BLOCK_WISE_N);\n- lU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(), getRowsInBlock(), getColsInBlock(), U.getNnz());\n- setLineNumbers(lU);\n- }\n- }\n- else {\n- //replication of U for shuffle to target block\n- Lop offset = createOffsetLop(V, false); //ncol of t(V) -> nrow of V determines num replicates\n- lU = new RepMat(U.constructLops(), offset, true, V.getDataType(), V.getValueType());\n- lU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(),\n- U.getRowsInBlock(), U.getColsInBlock(), U.getNnz());\n- setLineNumbers(lU);\n-\n- Group grpU = new Group(lU, Group.OperationTypes.Sort, DataType.MATRIX, ValueType.DOUBLE);\n- grpU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(), U.getRowsInBlock(), U.getColsInBlock(), -1);\n- setLineNumbers(grpU);\n- lU = grpU;\n- }\n-\n- Lop lV = null;\n- if( cacheV ) {\n- //partitioning of V for read through distributed cache\n- boolean needPartV = !V.dimsKnown() || V.getDim1() * V.getDim2() > DistributedCacheInput.PARTITION_SIZE;\n- lV = V.constructLops();\n- if( needPartV ){ //requires partitioning\n- lV = new DataPartition(lV, DataType.MATRIX, ValueType.DOUBLE, (m2Size>OptimizerUtils.getLocalMemBudget())?ExecType.MR:ExecType.CP, PDataPartitionFormat.ROW_BLOCK_WISE_N);\n- lV.getOutputParameters().setDimensions(V.getDim1(), V.getDim2(), getRowsInBlock(), getColsInBlock(), V.getNnz());\n- setLineNumbers(lV);\n- }\n- }\n- else {\n- //replication of t(V) for shuffle to target block\n- Transform ltV = new Transform( V.constructLops(), HopsTransf2Lops.get(ReOrgOp.TRANSPOSE), getDataType(), getValueType(), ExecType.MR);\n- ltV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(),\n- V.getColsInBlock(), V.getRowsInBlock(), V.getNnz());\n- setLineNumbers(ltV);\n-\n- Lop offset = createOffsetLop(U, false); //nrow of U determines num replicates\n- lV = new RepMat(ltV, offset, false, V.getDataType(), V.getValueType());\n- lV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(),\n- V.getColsInBlock(), V.getRowsInBlock(), V.getNnz());\n- setLineNumbers(lV);\n-\n- Group grpV = new Group(lV, Group.OperationTypes.Sort, DataType.MATRIX, ValueType.DOUBLE);\n- grpV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(), V.getColsInBlock(), V.getRowsInBlock(), -1);\n- setLineNumbers(grpV);\n- lV = grpV;\n- }\n+ Lop lU = obtainlU(U, V, cacheU, m1Size);\n+ Lop lV = obtainlV(U, V, cacheV, m2Size);\n//reduce-side wdivmm w/ or without broadcast\nLop wdivmm = new WeightedDivMMR( grpW, lU, lV, grpX,\n@@ -1007,60 +919,8 @@ public class QuaternaryOp extends Hop implements MultiThreadedHop\ngrpX.getOutputParameters().setDimensions(X.getDim1(), X.getDim2(), X.getRowsInBlock(), X.getColsInBlock(), -1);\nsetLineNumbers(grpX);\n- Lop lU = null;\n- if( cacheU ) {\n- //partitioning of U for read through distributed cache\n- boolean needPartU = !U.dimsKnown() || U.getDim1() * U.getDim2() > DistributedCacheInput.PARTITION_SIZE;\n- lU = U.constructLops();\n- if( needPartU ){ //requires partitioning\n- lU = new DataPartition(lU, DataType.MATRIX, ValueType.DOUBLE, (m1Size>OptimizerUtils.getLocalMemBudget())?ExecType.MR:ExecType.CP, PDataPartitionFormat.ROW_BLOCK_WISE_N);\n- lU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(), getRowsInBlock(), getColsInBlock(), U.getNnz());\n- setLineNumbers(lU);\n- }\n- }\n- else {\n- //replication of U for shuffle to target block\n- Lop offset = createOffsetLop(V, false); //ncol of t(V) -> nrow of V determines num replicates\n- lU = new RepMat(U.constructLops(), offset, true, V.getDataType(), V.getValueType());\n- lU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(),\n- U.getRowsInBlock(), U.getColsInBlock(), U.getNnz());\n- setLineNumbers(lU);\n-\n- Group grpU = new Group(lU, Group.OperationTypes.Sort, DataType.MATRIX, ValueType.DOUBLE);\n- grpU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(), U.getRowsInBlock(), U.getColsInBlock(), -1);\n- setLineNumbers(grpU);\n- lU = grpU;\n- }\n-\n- Lop lV = null;\n- if( cacheV ) {\n- //partitioning of V for read through distributed cache\n- boolean needPartV = !V.dimsKnown() || V.getDim1() * V.getDim2() > DistributedCacheInput.PARTITION_SIZE;\n- lV = V.constructLops();\n- if( needPartV ){ //requires partitioning\n- lV = new DataPartition(lV, DataType.MATRIX, ValueType.DOUBLE, (m2Size>OptimizerUtils.getLocalMemBudget())?ExecType.MR:ExecType.CP, PDataPartitionFormat.ROW_BLOCK_WISE_N);\n- lV.getOutputParameters().setDimensions(V.getDim1(), V.getDim2(), getRowsInBlock(), getColsInBlock(), V.getNnz());\n- setLineNumbers(lV);\n- }\n- }\n- else {\n- //replication of t(V) for shuffle to target block\n- Transform ltV = new Transform( V.constructLops(), HopsTransf2Lops.get(ReOrgOp.TRANSPOSE), getDataType(), getValueType(), ExecType.MR);\n- ltV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(),\n- V.getColsInBlock(), V.getRowsInBlock(), V.getNnz());\n- setLineNumbers(ltV);\n-\n- Lop offset = createOffsetLop(U, false); //nrow of U determines num replicates\n- lV = new RepMat(ltV, offset, false, V.getDataType(), V.getValueType());\n- lV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(),\n- V.getColsInBlock(), V.getRowsInBlock(), V.getNnz());\n- setLineNumbers(lV);\n-\n- Group grpV = new Group(lV, Group.OperationTypes.Sort, DataType.MATRIX, ValueType.DOUBLE);\n- grpV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(), V.getColsInBlock(), V.getRowsInBlock(), -1);\n- setLineNumbers(grpV);\n- lV = grpV;\n- }\n+ Lop lU = obtainlU(U, V, cacheU, m1Size);\n+ Lop lV = obtainlV(U, V, cacheV, m2Size);\n//reduce-side wcemm w/ or without broadcast\nLop wcemm = new WeightedCrossEntropyR( grpX, lU, lV, eps.constructLops(),\n@@ -1216,60 +1076,8 @@ public class QuaternaryOp extends Hop implements MultiThreadedHop\ngrpX.getOutputParameters().setDimensions(X.getDim1(), X.getDim2(), X.getRowsInBlock(), X.getColsInBlock(), X.getNnz());\nsetLineNumbers(grpX);\n- Lop lU = null;\n- if( cacheU ) {\n- //partitioning of U for read through distributed cache\n- boolean needPartU = !U.dimsKnown() || U.getDim1() * U.getDim2() > DistributedCacheInput.PARTITION_SIZE;\n- lU = U.constructLops();\n- if( needPartU ){ //requires partitioning\n- lU = new DataPartition(lU, DataType.MATRIX, ValueType.DOUBLE, (m1Size>OptimizerUtils.getLocalMemBudget())?ExecType.MR:ExecType.CP, PDataPartitionFormat.ROW_BLOCK_WISE_N);\n- lU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(), getRowsInBlock(), getColsInBlock(), U.getNnz());\n- setLineNumbers(lU);\n- }\n- }\n- else {\n- //replication of U for shuffle to target block\n- Lop offset = createOffsetLop(V, false); //ncol of t(V) -> nrow of V determines num replicates\n- lU = new RepMat(U.constructLops(), offset, true, V.getDataType(), V.getValueType());\n- lU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(),\n- U.getRowsInBlock(), U.getColsInBlock(), U.getNnz());\n- setLineNumbers(lU);\n-\n- Group grpU = new Group(lU, Group.OperationTypes.Sort, DataType.MATRIX, ValueType.DOUBLE);\n- grpU.getOutputParameters().setDimensions(U.getDim1(), U.getDim2(), U.getRowsInBlock(), U.getColsInBlock(), -1);\n- setLineNumbers(grpU);\n- lU = grpU;\n- }\n-\n- Lop lV = null;\n- if( cacheV ) {\n- //partitioning of V for read through distributed cache\n- boolean needPartV = !V.dimsKnown() || V.getDim1() * V.getDim2() > DistributedCacheInput.PARTITION_SIZE;\n- lV = V.constructLops();\n- if( needPartV ){ //requires partitioning\n- lV = new DataPartition(lV, DataType.MATRIX, ValueType.DOUBLE, (m2Size>OptimizerUtils.getLocalMemBudget())?ExecType.MR:ExecType.CP, PDataPartitionFormat.ROW_BLOCK_WISE_N);\n- lV.getOutputParameters().setDimensions(V.getDim1(), V.getDim2(), getRowsInBlock(), getColsInBlock(), V.getNnz());\n- setLineNumbers(lV);\n- }\n- }\n- else {\n- //replication of t(V) for shuffle to target block\n- Transform ltV = new Transform( V.constructLops(), HopsTransf2Lops.get(ReOrgOp.TRANSPOSE), getDataType(), getValueType(), ExecType.MR);\n- ltV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(),\n- V.getColsInBlock(), V.getRowsInBlock(), V.getNnz());\n- setLineNumbers(ltV);\n-\n- Lop offset = createOffsetLop(U, false); //nrow of U determines num replicates\n- lV = new RepMat(ltV, offset, false, V.getDataType(), V.getValueType());\n- lV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(),\n- V.getColsInBlock(), V.getRowsInBlock(), V.getNnz());\n- setLineNumbers(lV);\n-\n- Group grpV = new Group(lV, Group.OperationTypes.Sort, DataType.MATRIX, ValueType.DOUBLE);\n- grpV.getOutputParameters().setDimensions(V.getDim2(), V.getDim1(), V.getColsInBlock(), V.getRowsInBlock(), -1);\n- setLineNumbers(grpV);\n- lV = grpV;\n- }\n+ Lop lU = obtainlU(U, V, cacheU, m1Size);\n+ Lop lV = obtainlV(U, V, cacheV, m2Size);\n//reduce-side wumm w/ or without broadcast\nLop wumm = new WeightedUnaryMMR(\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/mlcontext/FrameMetadata.java", "new_path": "src/main/java/org/apache/sysml/api/mlcontext/FrameMetadata.java", "diff": "package org.apache.sysml.api.mlcontext;\n-import org.apache.sysml.conf.ConfigurationManager;\nimport org.apache.sysml.runtime.matrix.MatrixCharacteristics;\n/**\n@@ -30,11 +29,6 @@ import org.apache.sysml.runtime.matrix.MatrixCharacteristics;\n*/\npublic class FrameMetadata extends Metadata {\n- private Long numRows = null;\n- private Long numColumns = null;\n- private Long numNonZeros = null;\n- private Integer numRowsPerBlock = null;\n- private Integer numColumnsPerBlock = null;\nprivate FrameFormat frameFormat;\nprivate FrameSchema frameSchema;\n@@ -522,139 +516,6 @@ public class FrameMetadata extends Metadata {\nthis.numColumnsPerBlock = matrixCharacteristics.getColsPerBlock();\n}\n- /**\n- * Obtain the number of rows\n- *\n- * @return the number of rows\n- */\n- public Long getNumRows() {\n- return numRows;\n- }\n-\n- /**\n- * Set the number of rows\n- *\n- * @param numRows\n- * the number of rows\n- */\n- public void setNumRows(Long numRows) {\n- this.numRows = numRows;\n- }\n-\n- /**\n- * Obtain the number of columns\n- *\n- * @return the number of columns\n- */\n- public Long getNumColumns() {\n- return numColumns;\n- }\n-\n- /**\n- * Set the number of columns\n- *\n- * @param numColumns\n- * the number of columns\n- */\n- public void setNumColumns(Long numColumns) {\n- this.numColumns = numColumns;\n- }\n-\n- /**\n- * Obtain the number of non-zero values\n- *\n- * @return the number of non-zero values\n- */\n- public Long getNumNonZeros() {\n- return numNonZeros;\n- }\n-\n- /**\n- * Set the number of non-zero values\n- *\n- * @param numNonZeros\n- * the number of non-zero values\n- */\n- public void setNumNonZeros(Long numNonZeros) {\n- this.numNonZeros = numNonZeros;\n- }\n-\n- /**\n- * Obtain the number of rows per block\n- *\n- * @return the number of rows per block\n- */\n- public Integer getNumRowsPerBlock() {\n- return numRowsPerBlock;\n- }\n-\n- /**\n- * Set the number of rows per block\n- *\n- * @param numRowsPerBlock\n- * the number of rows per block\n- */\n- public void setNumRowsPerBlock(Integer numRowsPerBlock) {\n- this.numRowsPerBlock = numRowsPerBlock;\n- }\n-\n- /**\n- * Obtain the number of columns per block\n- *\n- * @return the number of columns per block\n- */\n- public Integer getNumColumnsPerBlock() {\n- return numColumnsPerBlock;\n- }\n-\n- /**\n- * Set the number of columns per block\n- *\n- * @param numColumnsPerBlock\n- * the number of columns per block\n- */\n- public void setNumColumnsPerBlock(Integer numColumnsPerBlock) {\n- this.numColumnsPerBlock = numColumnsPerBlock;\n- }\n-\n- /**\n- * Convert the frame metadata to a MatrixCharacteristics object. If all\n- * field values are {@code null}, {@code null} is returned.\n- *\n- * @return the frame metadata as a MatrixCharacteristics object, or\n- * {@code null} if all field values are null\n- */\n- public MatrixCharacteristics asMatrixCharacteristics() {\n-\n- if ((numRows == null) && (numColumns == null) && (numRowsPerBlock == null) && (numColumnsPerBlock == null)\n- && (numNonZeros == null)) {\n- return null;\n- }\n-\n- long nr = (numRows == null) ? -1 : numRows;\n- long nc = (numColumns == null) ? -1 : numColumns;\n- int nrpb = (numRowsPerBlock == null) ? ConfigurationManager.getBlocksize() : numRowsPerBlock;\n- int ncpb = (numColumnsPerBlock == null) ? ConfigurationManager.getBlocksize() : numColumnsPerBlock;\n- long nnz = (numNonZeros == null) ? -1 : numNonZeros;\n- MatrixCharacteristics mc = new MatrixCharacteristics(nr, nc, nrpb, ncpb, nnz);\n- return mc;\n- }\n-\n- @Override\n- public String toString() {\n- return \"rows: \" + fieldDisplay(numRows) + \", columns: \" + fieldDisplay(numColumns) + \", non-zeros: \"\n- + fieldDisplay(numNonZeros) + \", rows per block: \" + fieldDisplay(numRowsPerBlock)\n- + \", columns per block: \" + fieldDisplay(numColumnsPerBlock);\n- }\n-\n- private String fieldDisplay(Object field) {\n- if (field == null) {\n- return \"None\";\n- } else {\n- return field.toString();\n- }\n- }\n-\n/**\n* Obtain the frame format\n*\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/mlcontext/MatrixMetadata.java", "new_path": "src/main/java/org/apache/sysml/api/mlcontext/MatrixMetadata.java", "diff": "package org.apache.sysml.api.mlcontext;\n-import org.apache.sysml.conf.ConfigurationManager;\nimport org.apache.sysml.runtime.matrix.MatrixCharacteristics;\n/**\n@@ -30,11 +29,6 @@ import org.apache.sysml.runtime.matrix.MatrixCharacteristics;\n*/\npublic class MatrixMetadata extends Metadata {\n- private Long numRows = null;\n- private Long numColumns = null;\n- private Long numNonZeros = null;\n- private Integer numRowsPerBlock = null;\n- private Integer numColumnsPerBlock = null;\nprivate MatrixFormat matrixFormat;\npublic MatrixMetadata() {\n@@ -354,152 +348,6 @@ public class MatrixMetadata extends Metadata {\nthis.numColumnsPerBlock = matrixCharacteristics.getColsPerBlock();\n}\n- /**\n- * Set the MatrixMetadata fields based on a MatrixCharacteristics object.\n- *\n- * @param matrixCharacteristics\n- * the matrix metadata as a MatrixCharacteristics object\n- */\n- public void setMatrixCharacteristics(MatrixCharacteristics matrixCharacteristics) {\n- this.numRows = matrixCharacteristics.getRows();\n- this.numColumns = matrixCharacteristics.getCols();\n- this.numNonZeros = matrixCharacteristics.getNonZeros();\n- this.numRowsPerBlock = matrixCharacteristics.getRowsPerBlock();\n- this.numColumnsPerBlock = matrixCharacteristics.getColsPerBlock();\n- }\n-\n- /**\n- * Obtain the number of rows\n- *\n- * @return the number of rows\n- */\n- public Long getNumRows() {\n- return numRows;\n- }\n-\n- /**\n- * Set the number of rows\n- *\n- * @param numRows\n- * the number of rows\n- */\n- public void setNumRows(Long numRows) {\n- this.numRows = numRows;\n- }\n-\n- /**\n- * Obtain the number of columns\n- *\n- * @return the number of columns\n- */\n- public Long getNumColumns() {\n- return numColumns;\n- }\n-\n- /**\n- * Set the number of columns\n- *\n- * @param numColumns\n- * the number of columns\n- */\n- public void setNumColumns(Long numColumns) {\n- this.numColumns = numColumns;\n- }\n-\n- /**\n- * Obtain the number of non-zero values\n- *\n- * @return the number of non-zero values\n- */\n- public Long getNumNonZeros() {\n- return numNonZeros;\n- }\n-\n- /**\n- * Set the number of non-zero values\n- *\n- * @param numNonZeros\n- * the number of non-zero values\n- */\n- public void setNumNonZeros(Long numNonZeros) {\n- this.numNonZeros = numNonZeros;\n- }\n-\n- /**\n- * Obtain the number of rows per block\n- *\n- * @return the number of rows per block\n- */\n- public Integer getNumRowsPerBlock() {\n- return numRowsPerBlock;\n- }\n-\n- /**\n- * Set the number of rows per block\n- *\n- * @param numRowsPerBlock\n- * the number of rows per block\n- */\n- public void setNumRowsPerBlock(Integer numRowsPerBlock) {\n- this.numRowsPerBlock = numRowsPerBlock;\n- }\n-\n- /**\n- * Obtain the number of columns per block\n- *\n- * @return the number of columns per block\n- */\n- public Integer getNumColumnsPerBlock() {\n- return numColumnsPerBlock;\n- }\n-\n- /**\n- * Set the number of columns per block\n- *\n- * @param numColumnsPerBlock\n- * the number of columns per block\n- */\n- public void setNumColumnsPerBlock(Integer numColumnsPerBlock) {\n- this.numColumnsPerBlock = numColumnsPerBlock;\n- }\n-\n- /**\n- * Convert the matrix metadata to a MatrixCharacteristics object. If all\n- * field values are {@code null}, {@code null} is returned.\n- *\n- * @return the matrix metadata as a MatrixCharacteristics object, or\n- * {@code null} if all field values are null\n- */\n- public MatrixCharacteristics asMatrixCharacteristics() {\n-\n- if ((numRows == null) && (numColumns == null) && (numRowsPerBlock == null) && (numColumnsPerBlock == null)\n- && (numNonZeros == null)) {\n- return null;\n- }\n-\n- long nr = (numRows == null) ? -1 : numRows;\n- long nc = (numColumns == null) ? -1 : numColumns;\n- int nrpb = (numRowsPerBlock == null) ? ConfigurationManager.getBlocksize() : numRowsPerBlock;\n- int ncpb = (numColumnsPerBlock == null) ? ConfigurationManager.getBlocksize() : numColumnsPerBlock;\n- long nnz = (numNonZeros == null) ? -1 : numNonZeros;\n- return new MatrixCharacteristics(nr, nc, nrpb, ncpb, nnz);\n- }\n-\n- @Override\n- public String toString() {\n- return \"rows: \" + fieldDisplay(numRows) + \", columns: \" + fieldDisplay(numColumns) + \", non-zeros: \"\n- + fieldDisplay(numNonZeros) + \", rows per block: \" + fieldDisplay(numRowsPerBlock)\n- + \", columns per block: \" + fieldDisplay(numColumnsPerBlock);\n- }\n-\n- private String fieldDisplay(Object field) {\n- if (field == null) {\n- return \"None\";\n- } else {\n- return field.toString();\n- }\n- }\n-\n/**\n* Obtain the matrix format\n*\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/mlcontext/Metadata.java", "new_path": "src/main/java/org/apache/sysml/api/mlcontext/Metadata.java", "diff": "package org.apache.sysml.api.mlcontext;\n+import org.apache.sysml.conf.ConfigurationManager;\n+import org.apache.sysml.runtime.matrix.MatrixCharacteristics;\n+\n/**\n* Abstract metadata class for MLContext API. Complex types such as SystemML\n* matrices and frames typically require metadata, so this abstract class serves\n@@ -27,4 +30,156 @@ package org.apache.sysml.api.mlcontext;\n*/\npublic abstract class Metadata {\n+ protected Long numColumns = null;\n+ protected Integer numColumnsPerBlock = null;\n+ protected Long numNonZeros = null;\n+ protected Long numRows = null;\n+ protected Integer numRowsPerBlock = null;\n+\n+ /**\n+ * Convert the metadata to a MatrixCharacteristics object. If all field\n+ * values are {@code null}, {@code null} is returned.\n+ *\n+ * @return the metadata as a MatrixCharacteristics object, or {@code null}\n+ * if all field values are null\n+ */\n+ public MatrixCharacteristics asMatrixCharacteristics() {\n+\n+ if ((numRows == null) && (numColumns == null) && (numRowsPerBlock == null) && (numColumnsPerBlock == null)\n+ && (numNonZeros == null)) {\n+ return null;\n+ }\n+\n+ long nr = (numRows == null) ? -1 : numRows;\n+ long nc = (numColumns == null) ? -1 : numColumns;\n+ int nrpb = (numRowsPerBlock == null) ? ConfigurationManager.getBlocksize() : numRowsPerBlock;\n+ int ncpb = (numColumnsPerBlock == null) ? ConfigurationManager.getBlocksize() : numColumnsPerBlock;\n+ long nnz = (numNonZeros == null) ? -1 : numNonZeros;\n+ return new MatrixCharacteristics(nr, nc, nrpb, ncpb, nnz);\n+ }\n+\n+ protected String fieldDisplay(Object field) {\n+ if (field == null) {\n+ return \"None\";\n+ } else {\n+ return field.toString();\n+ }\n+ }\n+\n+ /**\n+ * Obtain the number of columns\n+ *\n+ * @return the number of columns\n+ */\n+ public Long getNumColumns() {\n+ return numColumns;\n+ }\n+\n+ /**\n+ * Obtain the number of columns per block\n+ *\n+ * @return the number of columns per block\n+ */\n+ public Integer getNumColumnsPerBlock() {\n+ return numColumnsPerBlock;\n+ }\n+\n+ /**\n+ * Obtain the number of non-zero values\n+ *\n+ * @return the number of non-zero values\n+ */\n+ public Long getNumNonZeros() {\n+ return numNonZeros;\n+ }\n+\n+ /**\n+ * Obtain the number of rows\n+ *\n+ * @return the number of rows\n+ */\n+ public Long getNumRows() {\n+ return numRows;\n+ }\n+\n+ /**\n+ * Obtain the number of rows per block\n+ *\n+ * @return the number of rows per block\n+ */\n+ public Integer getNumRowsPerBlock() {\n+ return numRowsPerBlock;\n+ }\n+\n+ /**\n+ * Set the metadata fields based on a MatrixCharacteristics object.\n+ *\n+ * @param matrixCharacteristics\n+ * the matrix metadata as a MatrixCharacteristics object\n+ */\n+ public void setMatrixCharacteristics(MatrixCharacteristics matrixCharacteristics) {\n+ this.numRows = matrixCharacteristics.getRows();\n+ this.numColumns = matrixCharacteristics.getCols();\n+ this.numNonZeros = matrixCharacteristics.getNonZeros();\n+ this.numRowsPerBlock = matrixCharacteristics.getRowsPerBlock();\n+ this.numColumnsPerBlock = matrixCharacteristics.getColsPerBlock();\n+ }\n+\n+ /**\n+ * Set the number of columns\n+ *\n+ * @param numColumns\n+ * the number of columns\n+ */\n+ public void setNumColumns(Long numColumns) {\n+ this.numColumns = numColumns;\n+ }\n+\n+ /**\n+ * Set the number of columns per block\n+ *\n+ * @param numColumnsPerBlock\n+ * the number of columns per block\n+ */\n+ public void setNumColumnsPerBlock(Integer numColumnsPerBlock) {\n+ this.numColumnsPerBlock = numColumnsPerBlock;\n+ }\n+\n+ /**\n+ * Set the number of non-zero values\n+ *\n+ * @param numNonZeros\n+ * the number of non-zero values\n+ */\n+ public void setNumNonZeros(Long numNonZeros) {\n+ this.numNonZeros = numNonZeros;\n+ }\n+\n+ /**\n+ * Set the number of rows\n+ *\n+ * @param numRows\n+ * the number of rows\n+ */\n+ public void setNumRows(Long numRows) {\n+ this.numRows = numRows;\n+ }\n+\n+ /**\n+ * Set the number of rows per block\n+ *\n+ * @param numRowsPerBlock\n+ * the number of rows per block\n+ */\n+ public void setNumRowsPerBlock(Integer numRowsPerBlock) {\n+ this.numRowsPerBlock = numRowsPerBlock;\n+ }\n+\n+ @Override\n+ public String toString() {\n+ return \"rows: \" + fieldDisplay(numRows) + \", columns: \" + fieldDisplay(numColumns) + \", non-zeros: \"\n+ + fieldDisplay(numNonZeros) + \", rows per block: \" + fieldDisplay(numRowsPerBlock)\n+ + \", columns per block: \" + fieldDisplay(numColumnsPerBlock);\n+ }\n+\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "docs/python-performance-test.md", "new_path": "docs/python-performance-test.md", "diff": "@@ -148,6 +148,27 @@ Run performance test for all algorithms under the family `regression2` and log w\nRun performance test for all algorithms using HDFS.\n+## Google sheets API\n+\n+Steps below to configure google client API:\n+\n+- Navigate to [Google APIs Console](https://console.developers.google.com/apis/).\n+- Create a new project.\n+- Click Enable API. Search for and enable the Google Drive API.\n+- Create credentials for a Web Server to access Application Data.\n+- Name the service account and grant it a Project Role of Editor.\n+- Download the JSON file.\n+- Copy the JSON file to your code directory and rename it to client_secret.json\n+\n+Steps below to configure google sheets:\n+\n+- Create a new spread sheet with google sheets.\n+- Create seperate sheets for `singlenode` and `hybrid_spark`.\n+- Find the client_email inside client_secret.json and save it.\n+- Back in your spreadsheet, click the Share button in the top right, and paste the client email into the People field to give it edit rights for each sheet.\n+- Click Send\n+\n+\n## Result Consolidation and Plotting\nWe have two scripts, `stats.py` forpulling results from google docs and `update.py` to updating results to google docs or local file system.\n@@ -159,6 +180,7 @@ Example of `stats.py` below\n` ./stats.py --auth ../key/client_json.json --exec-type singlenode --plot stats1_data-gen_none_dense_10k_100`\n`--plot` argument needs the name of the composite key that you would like to compare results over. If this argument is not specified the results would be grouped by keys.\n+\n## Operational Notes\nAll performance test depend mainly on two scripts for execution `systemml-standalone.py` and `systemml-spark-submit.py`. Incase we need to change standalone or spark parameters we need to manually change these parameters in their respective scripts.\n@@ -198,7 +220,6 @@ Matrix Shape | Approximate Data Size\n10M_1k|80GB\n100M_1k|800GB\n-\nFor example the command below runs performance test for all data sizes described above\n`run_perftest.py --family binomial clustering multinomial regression1 regression2 stats1 stats2 --mat-shape 10k_1k 100k_1k 1M_1k 10M_1k 100M_1k --master yarn-client --temp-dir hdfs://localhost:9000/user/systemml`\n" } ]

[ { "change_type": "MODIFY", "old_path": "dev/release/src/test/java/org/apache/sysml/validation/ValidateLicAndNotice.java", "new_path": "dev/release/src/test/java/org/apache/sysml/validation/ValidateLicAndNotice.java", "diff": "@@ -59,7 +59,10 @@ 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+ {\"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};\npublic static HashMap<String, String[][]> hmJSLicenses = new HashMap<String, String[][]>();\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/systemml/mllearn/estimators.py", "new_path": "src/main/python/systemml/mllearn/estimators.py", "diff": "@@ -100,6 +100,19 @@ class BaseSystemMLEstimator(Estimator):\nself.estimator.setExplain(explain)\nreturn self\n+ def setExplainLevel(self, explainLevel):\n+ \"\"\"\n+ Set explain level. Mainly intended for developers.\n+\n+ Parameters\n+ ----------\n+ explainLevel: string\n+ Can be one of \"hops\", \"runtime\", \"recompile_hops\", \"recompile_runtime\"\n+ or in the above in upper case.\n+ \"\"\"\n+ self.estimator.setExplainLevel(explainLevel)\n+ return self\n+\ndef setStatistics(self, statistics):\n\"\"\"\nWhether or not to output statistics (such as execution time, elapsed time)\n" }, { "change_type": "MODIFY", "old_path": "src/main/scala/org/apache/sysml/api/ml/BaseSystemMLClassifier.scala", "new_path": "src/main/scala/org/apache/sysml/api/ml/BaseSystemMLClassifier.scala", "diff": "@@ -108,23 +108,27 @@ trait BaseSystemMLEstimatorOrModel {\nvar enableGPU:Boolean = false\nvar forceGPU:Boolean = false\nvar explain:Boolean = false\n+ var explainLevel:String = \"runtime\"\nvar statistics:Boolean = false\nvar statisticsMaxHeavyHitters:Int = 10\nval config:HashMap[String, String] = new HashMap[String, String]()\ndef setGPU(enableGPU1:Boolean):BaseSystemMLEstimatorOrModel = { enableGPU = enableGPU1; this}\ndef setForceGPU(enableGPU1:Boolean):BaseSystemMLEstimatorOrModel = { forceGPU = enableGPU1; this}\ndef setExplain(explain1:Boolean):BaseSystemMLEstimatorOrModel = { explain = explain1; this}\n+ def setExplainLevel(explainLevel1:String):BaseSystemMLEstimatorOrModel = { explainLevel = explainLevel1; this }\ndef setStatistics(statistics1:Boolean):BaseSystemMLEstimatorOrModel = { statistics = statistics1; this}\ndef setStatisticsMaxHeavyHitters(statisticsMaxHeavyHitters1:Int):BaseSystemMLEstimatorOrModel = { statisticsMaxHeavyHitters = statisticsMaxHeavyHitters1; this}\ndef setConfigProperty(key:String, value:String):BaseSystemMLEstimatorOrModel = { config.put(key, value); this}\ndef updateML(ml:MLContext):Unit = {\nml.setGPU(enableGPU); ml.setForceGPU(forceGPU);\n- ml.setExplain(explain); ml.setStatistics(statistics); ml.setStatisticsMaxHeavyHitters(statisticsMaxHeavyHitters);\n+ ml.setExplain(explain); ml.setExplainLevel(explainLevel);\n+ ml.setStatistics(statistics); ml.setStatisticsMaxHeavyHitters(statisticsMaxHeavyHitters);\nconfig.map(x => ml.setConfigProperty(x._1, x._2))\n}\ndef copyProperties(other:BaseSystemMLEstimatorOrModel):BaseSystemMLEstimatorOrModel = {\nother.setGPU(enableGPU); other.setForceGPU(forceGPU);\n- other.setExplain(explain); other.setStatistics(statistics); other.setStatisticsMaxHeavyHitters(statisticsMaxHeavyHitters);\n+ other.setExplain(explain); other.setExplainLevel(explainLevel);\n+ other.setStatistics(statistics); other.setStatisticsMaxHeavyHitters(statisticsMaxHeavyHitters);\nconfig.map(x => other.setConfigProperty(x._1, x._2))\nreturn other\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/mlcontext/MLContext.java", "new_path": "src/main/java/org/apache/sysml/api/mlcontext/MLContext.java", "diff": "@@ -55,7 +55,7 @@ public class MLContext {\n/**\n* Logger for MLContext\n*/\n- public static Logger log = Logger.getLogger(MLContext.class);\n+ protected static Logger log = Logger.getLogger(MLContext.class);\n/**\n* SparkSession object.\n@@ -665,7 +665,9 @@ public class MLContext {\n// clear local status, but do not stop sc as it\n// may be used or stopped externally\n+ if (executionScript != null) {\nexecutionScript.clearAll();\n+ }\nresetConfig();\nspark = null;\n}\n@@ -693,7 +695,7 @@ public class MLContext {\n*/\npublic String version() {\nif (info() == null) {\n- return \"Version not available\";\n+ return MLContextUtil.VERSION_NOT_AVAILABLE;\n}\nreturn info().version();\n}\n@@ -705,7 +707,7 @@ public class MLContext {\n*/\npublic String buildTime() {\nif (info() == null) {\n- return \"Build time not available\";\n+ return MLContextUtil.BUILD_TIME_NOT_AVAILABLE;\n}\nreturn info().buildTime();\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/mlcontext/MLContextConversionUtil.java", "new_path": "src/main/java/org/apache/sysml/api/mlcontext/MLContextConversionUtil.java", "diff": "@@ -482,6 +482,9 @@ public class MLContextConversionUtil {\n* the matrix metadata, if available\n*/\npublic static void determineMatrixFormatIfNeeded(Dataset<Row> dataFrame, MatrixMetadata matrixMetadata) {\n+ if (matrixMetadata == null) {\n+ return;\n+ }\nMatrixFormat matrixFormat = matrixMetadata.getMatrixFormat();\nif (matrixFormat != null) {\nreturn;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/mlcontext/MLContextUtil.java", "new_path": "src/main/java/org/apache/sysml/api/mlcontext/MLContextUtil.java", "diff": "@@ -93,120 +93,30 @@ import org.w3c.dom.NodeList;\npublic final class MLContextUtil {\n/**\n- * Get HOP DAG in dot format for a DML or PYDML Script.\n- *\n- * @param mlCtx\n- * MLContext object.\n- * @param script\n- * The DML or PYDML Script object to execute.\n- * @param lines\n- * Only display the hops that have begin and end line number\n- * equals to the given integers.\n- * @param performHOPRewrites\n- * should perform static rewrites, perform\n- * intra-/inter-procedural analysis to propagate size information\n- * into functions and apply dynamic rewrites\n- * @param withSubgraph\n- * If false, the dot graph will be created without subgraphs for\n- * statement blocks.\n- * @return hop DAG in dot format\n- * @throws LanguageException\n- * if error occurs\n- * @throws DMLRuntimeException\n- * if error occurs\n- * @throws HopsException\n- * if error occurs\n+ * Version not available message.\n*/\n- public static String getHopDAG(MLContext mlCtx, Script script, ArrayList<Integer> lines,\n- boolean performHOPRewrites, boolean withSubgraph) throws HopsException, DMLRuntimeException,\n- LanguageException {\n- return getHopDAG(mlCtx, script, lines, null, performHOPRewrites, withSubgraph);\n- }\n+ public static final String VERSION_NOT_AVAILABLE = \"Version not available\";\n/**\n- * Get HOP DAG in dot format for a DML or PYDML Script.\n- *\n- * @param mlCtx\n- * MLContext object.\n- * @param script\n- * The DML or PYDML Script object to execute.\n- * @param lines\n- * Only display the hops that have begin and end line number\n- * equals to the given integers.\n- * @param newConf\n- * Spark Configuration.\n- * @param performHOPRewrites\n- * should perform static rewrites, perform\n- * intra-/inter-procedural analysis to propagate size information\n- * into functions and apply dynamic rewrites\n- * @param withSubgraph\n- * If false, the dot graph will be created without subgraphs for\n- * statement blocks.\n- * @return hop DAG in dot format\n- * @throws LanguageException\n- * if error occurs\n- * @throws DMLRuntimeException\n- * if error occurs\n- * @throws HopsException\n- * if error occurs\n+ * Build time not available message.\n*/\n- public static String getHopDAG(MLContext mlCtx, Script script, ArrayList<Integer> lines, SparkConf newConf,\n- boolean performHOPRewrites, boolean withSubgraph) throws HopsException, DMLRuntimeException,\n- LanguageException {\n- SparkConf oldConf = mlCtx.getSparkSession().sparkContext().getConf();\n- SparkExecutionContext.SparkClusterConfig systemmlConf = SparkExecutionContext.getSparkClusterConfig();\n- long oldMaxMemory = InfrastructureAnalyzer.getLocalMaxMemory();\n- try {\n- if (newConf != null) {\n- systemmlConf.analyzeSparkConfiguation(newConf);\n- InfrastructureAnalyzer.setLocalMaxMemory(newConf.getSizeAsBytes(\"spark.driver.memory\"));\n- }\n- ScriptExecutor scriptExecutor = new ScriptExecutor();\n- scriptExecutor.setExecutionType(mlCtx.getExecutionType());\n- scriptExecutor.setGPU(mlCtx.isGPU());\n- scriptExecutor.setForceGPU(mlCtx.isForceGPU());\n- scriptExecutor.setInit(mlCtx.isInitBeforeExecution());\n- if (mlCtx.isInitBeforeExecution()) {\n- mlCtx.setInitBeforeExecution(false);\n- }\n- scriptExecutor.setMaintainSymbolTable(mlCtx.isMaintainSymbolTable());\n-\n- Long time = new Long((new Date()).getTime());\n- if ((script.getName() == null) || (script.getName().equals(\"\"))) {\n- script.setName(time.toString());\n- }\n-\n- mlCtx.setExecutionScript(script);\n- scriptExecutor.compile(script, performHOPRewrites);\n- Explain.reset();\n- // To deal with potential Py4J issues\n- lines = lines.size() == 1 && lines.get(0) == -1 ? new ArrayList<Integer>() : lines;\n- return Explain.getHopDAG(scriptExecutor.dmlProgram, lines, withSubgraph);\n- } catch (RuntimeException e) {\n- throw new MLContextException(\"Exception when compiling script\", e);\n- } finally {\n- if (newConf != null) {\n- systemmlConf.analyzeSparkConfiguation(oldConf);\n- InfrastructureAnalyzer.setLocalMaxMemory(oldMaxMemory);\n- }\n- }\n- }\n+ public static final String BUILD_TIME_NOT_AVAILABLE = \"Build time not available\";\n/**\n- * Basic data types supported by the MLContext API\n+ * Basic data types supported by the MLContext API.\n*/\n@SuppressWarnings(\"rawtypes\")\npublic static final Class[] BASIC_DATA_TYPES = { Integer.class, Boolean.class, Double.class, String.class };\n/**\n- * Complex data types supported by the MLContext API\n+ * Complex data types supported by the MLContext API.\n*/\n@SuppressWarnings(\"rawtypes\")\npublic static final Class[] COMPLEX_DATA_TYPES = { JavaRDD.class, RDD.class, Dataset.class, Matrix.class,\nFrame.class, (new double[][] {}).getClass(), MatrixBlock.class, URL.class };\n/**\n- * All data types supported by the MLContext API\n+ * All data types supported by the MLContext API.\n*/\n@SuppressWarnings(\"rawtypes\")\npublic static final Class[] ALL_SUPPORTED_DATA_TYPES = (Class[]) ArrayUtils.addAll(BASIC_DATA_TYPES,\n@@ -1252,4 +1162,104 @@ public final class MLContextUtil {\n}\n}\n}\n+\n+ /**\n+ * Get HOP DAG in dot format for a DML or PYDML Script.\n+ *\n+ * @param mlCtx\n+ * MLContext object.\n+ * @param script\n+ * The DML or PYDML Script object to execute.\n+ * @param lines\n+ * Only display the hops that have begin and end line number\n+ * equals to the given integers.\n+ * @param performHOPRewrites\n+ * should perform static rewrites, perform\n+ * intra-/inter-procedural analysis to propagate size information\n+ * into functions and apply dynamic rewrites\n+ * @param withSubgraph\n+ * If false, the dot graph will be created without subgraphs for\n+ * statement blocks.\n+ * @return hop DAG in dot format\n+ * @throws LanguageException\n+ * if error occurs\n+ * @throws DMLRuntimeException\n+ * if error occurs\n+ * @throws HopsException\n+ * if error occurs\n+ */\n+ public static String getHopDAG(MLContext mlCtx, Script script, ArrayList<Integer> lines, boolean performHOPRewrites,\n+ boolean withSubgraph) throws HopsException, DMLRuntimeException, LanguageException {\n+ return getHopDAG(mlCtx, script, lines, null, performHOPRewrites, withSubgraph);\n+ }\n+\n+ /**\n+ * Get HOP DAG in dot format for a DML or PYDML Script.\n+ *\n+ * @param mlCtx\n+ * MLContext object.\n+ * @param script\n+ * The DML or PYDML Script object to execute.\n+ * @param lines\n+ * Only display the hops that have begin and end line number\n+ * equals to the given integers.\n+ * @param newConf\n+ * Spark Configuration.\n+ * @param performHOPRewrites\n+ * should perform static rewrites, perform\n+ * intra-/inter-procedural analysis to propagate size information\n+ * into functions and apply dynamic rewrites\n+ * @param withSubgraph\n+ * If false, the dot graph will be created without subgraphs for\n+ * statement blocks.\n+ * @return hop DAG in dot format\n+ * @throws LanguageException\n+ * if error occurs\n+ * @throws DMLRuntimeException\n+ * if error occurs\n+ * @throws HopsException\n+ * if error occurs\n+ */\n+ public static String getHopDAG(MLContext mlCtx, Script script, ArrayList<Integer> lines, SparkConf newConf,\n+ boolean performHOPRewrites, boolean withSubgraph)\n+ throws HopsException, DMLRuntimeException, LanguageException {\n+ SparkConf oldConf = mlCtx.getSparkSession().sparkContext().getConf();\n+ SparkExecutionContext.SparkClusterConfig systemmlConf = SparkExecutionContext.getSparkClusterConfig();\n+ long oldMaxMemory = InfrastructureAnalyzer.getLocalMaxMemory();\n+ try {\n+ if (newConf != null) {\n+ systemmlConf.analyzeSparkConfiguation(newConf);\n+ InfrastructureAnalyzer.setLocalMaxMemory(newConf.getSizeAsBytes(\"spark.driver.memory\"));\n+ }\n+ ScriptExecutor scriptExecutor = new ScriptExecutor();\n+ scriptExecutor.setExecutionType(mlCtx.getExecutionType());\n+ scriptExecutor.setGPU(mlCtx.isGPU());\n+ scriptExecutor.setForceGPU(mlCtx.isForceGPU());\n+ scriptExecutor.setInit(mlCtx.isInitBeforeExecution());\n+ if (mlCtx.isInitBeforeExecution()) {\n+ mlCtx.setInitBeforeExecution(false);\n+ }\n+ scriptExecutor.setMaintainSymbolTable(mlCtx.isMaintainSymbolTable());\n+\n+ Long time = new Long((new Date()).getTime());\n+ if ((script.getName() == null) || (script.getName().equals(\"\"))) {\n+ script.setName(time.toString());\n+ }\n+\n+ mlCtx.setExecutionScript(script);\n+ scriptExecutor.compile(script, performHOPRewrites);\n+ Explain.reset();\n+ // To deal with potential Py4J issues\n+ lines = lines.size() == 1 && lines.get(0) == -1 ? new ArrayList<Integer>() : lines;\n+ return Explain.getHopDAG(scriptExecutor.dmlProgram, lines, withSubgraph);\n+ } catch (RuntimeException e) {\n+ throw new MLContextException(\"Exception when compiling script\", e);\n+ } finally {\n+ if (newConf != null) {\n+ systemmlConf.analyzeSparkConfiguation(oldConf);\n+ InfrastructureAnalyzer.setLocalMaxMemory(oldMaxMemory);\n+ }\n+ }\n+ }\n+\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/mlcontext/MLContextTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/mlcontext/MLContextTest.java", "diff": "@@ -42,10 +42,13 @@ import java.util.Arrays;\nimport java.util.HashMap;\nimport java.util.List;\nimport java.util.Map;\n+import java.util.stream.Collectors;\n+import java.util.stream.Stream;\nimport org.apache.spark.api.java.JavaPairRDD;\nimport org.apache.spark.api.java.JavaRDD;\nimport org.apache.spark.api.java.function.Function;\n+import org.apache.spark.ml.linalg.DenseVector;\nimport org.apache.spark.ml.linalg.Vector;\nimport org.apache.spark.ml.linalg.VectorUDT;\nimport org.apache.spark.ml.linalg.Vectors;\n@@ -54,10 +57,12 @@ import org.apache.spark.sql.Dataset;\nimport org.apache.spark.sql.Row;\nimport org.apache.spark.sql.RowFactory;\nimport org.apache.spark.sql.types.DataTypes;\n+import org.apache.spark.sql.types.DoubleType;\nimport org.apache.spark.sql.types.StructField;\nimport org.apache.spark.sql.types.StructType;\nimport org.apache.sysml.api.mlcontext.MLContextConversionUtil;\nimport org.apache.sysml.api.mlcontext.MLContextException;\n+import org.apache.sysml.api.mlcontext.MLContextUtil;\nimport org.apache.sysml.api.mlcontext.MLResults;\nimport org.apache.sysml.api.mlcontext.Matrix;\nimport org.apache.sysml.api.mlcontext.MatrixFormat;\n@@ -69,11 +74,14 @@ import org.apache.sysml.runtime.instructions.spark.utils.RDDConverterUtils;\nimport org.apache.sysml.runtime.matrix.MatrixCharacteristics;\nimport org.apache.sysml.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysml.runtime.matrix.data.MatrixIndexes;\n+import org.apache.sysml.runtime.util.DataConverter;\nimport org.junit.Assert;\nimport org.junit.Test;\n+import scala.Tuple1;\nimport scala.Tuple2;\nimport scala.Tuple3;\n+import scala.Tuple4;\nimport scala.collection.Iterator;\nimport scala.collection.JavaConversions;\nimport scala.collection.Seq;\n@@ -2756,4 +2764,310 @@ public class MLContextTest extends MLContextTestBase {\nAssert.assertEquals(3, results.getLong(\"y\"));\n}\n+ @Test\n+ public void testOutputDataFrameOfVectorsDML() {\n+ System.out.println(\"MLContextTest - output DataFrame of vectors DML\");\n+\n+ String s = \"m=matrix('1 2 3 4',rows=2,cols=2);\";\n+ Script script = dml(s).out(\"m\");\n+ MLResults results = ml.execute(script);\n+ Dataset<Row> df = results.getDataFrame(\"m\", true);\n+ Dataset<Row> sortedDF = df.sort(RDDConverterUtils.DF_ID_COLUMN);\n+\n+ // verify column types\n+ StructType schema = sortedDF.schema();\n+ StructField[] fields = schema.fields();\n+ StructField idColumn = fields[0];\n+ StructField vectorColumn = fields[1];\n+ Assert.assertTrue(idColumn.dataType() instanceof DoubleType);\n+ Assert.assertTrue(vectorColumn.dataType() instanceof VectorUDT);\n+\n+ List<Row> list = sortedDF.collectAsList();\n+\n+ Row row1 = list.get(0);\n+ Assert.assertEquals(1.0, row1.getDouble(0), 0.0);\n+ Vector v1 = (DenseVector) row1.get(1);\n+ double[] arr1 = v1.toArray();\n+ Assert.assertArrayEquals(new double[] { 1.0, 2.0 }, arr1, 0.0);\n+\n+ Row row2 = list.get(1);\n+ Assert.assertEquals(2.0, row2.getDouble(0), 0.0);\n+ Vector v2 = (DenseVector) row2.get(1);\n+ double[] arr2 = v2.toArray();\n+ Assert.assertArrayEquals(new double[] { 3.0, 4.0 }, arr2, 0.0);\n+ }\n+\n+ @Test\n+ public void testOutputDoubleArrayFromMatrixDML() {\n+ System.out.println(\"MLContextTest - output double array from matrix DML\");\n+\n+ String s = \"M = matrix('1 2 3 4', rows=2, cols=2);\";\n+ double[][] matrix = ml.execute(dml(s).out(\"M\")).getMatrix(\"M\").to2DDoubleArray();\n+ Assert.assertEquals(1.0, matrix[0][0], 0);\n+ Assert.assertEquals(2.0, matrix[0][1], 0);\n+ Assert.assertEquals(3.0, matrix[1][0], 0);\n+ Assert.assertEquals(4.0, matrix[1][1], 0);\n+ }\n+\n+ @Test\n+ public void testOutputDataFrameFromMatrixDML() {\n+ System.out.println(\"MLContextTest - output DataFrame from matrix DML\");\n+\n+ String s = \"M = matrix('1 2 3 4', rows=2, cols=2);\";\n+ Script script = dml(s).out(\"M\");\n+ Dataset<Row> df = ml.execute(script).getMatrix(\"M\").toDF();\n+ Dataset<Row> sortedDF = df.sort(RDDConverterUtils.DF_ID_COLUMN);\n+ List<Row> list = sortedDF.collectAsList();\n+ Row row1 = list.get(0);\n+ Assert.assertEquals(1.0, row1.getDouble(0), 0.0);\n+ Assert.assertEquals(1.0, row1.getDouble(1), 0.0);\n+ Assert.assertEquals(2.0, row1.getDouble(2), 0.0);\n+\n+ Row row2 = list.get(1);\n+ Assert.assertEquals(2.0, row2.getDouble(0), 0.0);\n+ Assert.assertEquals(3.0, row2.getDouble(1), 0.0);\n+ Assert.assertEquals(4.0, row2.getDouble(2), 0.0);\n+ }\n+\n+ @Test\n+ public void testOutputDataFrameDoublesNoIDColumnFromMatrixDML() {\n+ System.out.println(\"MLContextTest - output DataFrame of doubles with no ID column from matrix DML\");\n+\n+ String s = \"M = matrix('1 2 3 4', rows=1, cols=4);\";\n+ Script script = dml(s).out(\"M\");\n+ Dataset<Row> df = ml.execute(script).getMatrix(\"M\").toDFDoubleNoIDColumn();\n+ List<Row> list = df.collectAsList();\n+\n+ Row row = list.get(0);\n+ Assert.assertEquals(1.0, row.getDouble(0), 0.0);\n+ Assert.assertEquals(2.0, row.getDouble(1), 0.0);\n+ Assert.assertEquals(3.0, row.getDouble(2), 0.0);\n+ Assert.assertEquals(4.0, row.getDouble(3), 0.0);\n+ }\n+\n+ @Test\n+ public void testOutputDataFrameDoublesWithIDColumnFromMatrixDML() {\n+ System.out.println(\"MLContextTest - output DataFrame of doubles with ID column from matrix DML\");\n+\n+ String s = \"M = matrix('1 2 3 4', rows=2, cols=2);\";\n+ Script script = dml(s).out(\"M\");\n+ Dataset<Row> df = ml.execute(script).getMatrix(\"M\").toDFDoubleWithIDColumn();\n+ Dataset<Row> sortedDF = df.sort(RDDConverterUtils.DF_ID_COLUMN);\n+ List<Row> list = sortedDF.collectAsList();\n+\n+ Row row1 = list.get(0);\n+ Assert.assertEquals(1.0, row1.getDouble(0), 0.0);\n+ Assert.assertEquals(1.0, row1.getDouble(1), 0.0);\n+ Assert.assertEquals(2.0, row1.getDouble(2), 0.0);\n+\n+ Row row2 = list.get(1);\n+ Assert.assertEquals(2.0, row2.getDouble(0), 0.0);\n+ Assert.assertEquals(3.0, row2.getDouble(1), 0.0);\n+ Assert.assertEquals(4.0, row2.getDouble(2), 0.0);\n+ }\n+\n+ @Test\n+ public void testOutputDataFrameVectorsNoIDColumnFromMatrixDML() {\n+ System.out.println(\"MLContextTest - output DataFrame of vectors with no ID column from matrix DML\");\n+\n+ String s = \"M = matrix('1 2 3 4', rows=1, cols=4);\";\n+ Script script = dml(s).out(\"M\");\n+ Dataset<Row> df = ml.execute(script).getMatrix(\"M\").toDFVectorNoIDColumn();\n+ List<Row> list = df.collectAsList();\n+\n+ Row row = list.get(0);\n+ Assert.assertArrayEquals(new double[] { 1.0, 2.0, 3.0, 4.0 }, ((Vector) row.get(0)).toArray(), 0.0);\n+ }\n+\n+ @Test\n+ public void testOutputDataFrameVectorsWithIDColumnFromMatrixDML() {\n+ System.out.println(\"MLContextTest - output DataFrame of vectors with ID column from matrix DML\");\n+\n+ String s = \"M = matrix('1 2 3 4', rows=1, cols=4);\";\n+ Script script = dml(s).out(\"M\");\n+ Dataset<Row> df = ml.execute(script).getMatrix(\"M\").toDFVectorWithIDColumn();\n+ List<Row> list = df.collectAsList();\n+\n+ Row row = list.get(0);\n+ Assert.assertEquals(1.0, row.getDouble(0), 0.0);\n+ Assert.assertArrayEquals(new double[] { 1.0, 2.0, 3.0, 4.0 }, ((Vector) row.get(1)).toArray(), 0.0);\n+ }\n+\n+ @Test\n+ public void testOutputJavaRDDStringCSVFromMatrixDML() {\n+ System.out.println(\"MLContextTest - output Java RDD String CSV from matrix DML\");\n+\n+ String s = \"M = matrix('1 2 3 4', rows=1, cols=4);\";\n+ Script script = dml(s).out(\"M\");\n+ JavaRDD<String> javaRDDStringCSV = ml.execute(script).getMatrix(\"M\").toJavaRDDStringCSV();\n+ List<String> lines = javaRDDStringCSV.collect();\n+ Assert.assertEquals(\"1.0,2.0,3.0,4.0\", lines.get(0));\n+ }\n+\n+ @Test\n+ public void testOutputJavaRDDStringIJVFromMatrixDML() {\n+ System.out.println(\"MLContextTest - output Java RDD String IJV from matrix DML\");\n+\n+ String s = \"M = matrix('1 2 3 4', rows=2, cols=2);\";\n+ Script script = dml(s).out(\"M\");\n+ MLResults results = ml.execute(script);\n+ JavaRDD<String> javaRDDStringIJV = results.getJavaRDDStringIJV(\"M\");\n+ List<String> lines = javaRDDStringIJV.sortBy(row -> row, true, 1).collect();\n+ Assert.assertEquals(\"1 1 1.0\", lines.get(0));\n+ Assert.assertEquals(\"1 2 2.0\", lines.get(1));\n+ Assert.assertEquals(\"2 1 3.0\", lines.get(2));\n+ Assert.assertEquals(\"2 2 4.0\", lines.get(3));\n+ }\n+\n+ @Test\n+ public void testOutputRDDStringCSVFromMatrixDML() {\n+ System.out.println(\"MLContextTest - output RDD String CSV from matrix DML\");\n+\n+ String s = \"M = matrix('1 2 3 4', rows=1, cols=4);\";\n+ Script script = dml(s).out(\"M\");\n+ RDD<String> rddStringCSV = ml.execute(script).getMatrix(\"M\").toRDDStringCSV();\n+ Iterator<String> iterator = rddStringCSV.toLocalIterator();\n+ Assert.assertEquals(\"1.0,2.0,3.0,4.0\", iterator.next());\n+ }\n+\n+ @Test\n+ public void testOutputRDDStringIJVFromMatrixDML() {\n+ System.out.println(\"MLContextTest - output RDD String IJV from matrix DML\");\n+\n+ String s = \"M = matrix('1 2 3 4', rows=2, cols=2);\";\n+ Script script = dml(s).out(\"M\");\n+ RDD<String> rddStringIJV = ml.execute(script).getMatrix(\"M\").toRDDStringIJV();\n+ String[] rows = (String[]) rddStringIJV.collect();\n+ Arrays.sort(rows);\n+ Assert.assertEquals(\"1 1 1.0\", rows[0]);\n+ Assert.assertEquals(\"1 2 2.0\", rows[1]);\n+ Assert.assertEquals(\"2 1 3.0\", rows[2]);\n+ Assert.assertEquals(\"2 2 4.0\", rows[3]);\n+ }\n+\n+ @Test\n+ public void testMLContextVersionMessage() {\n+ System.out.println(\"MLContextTest - version message\");\n+\n+ String version = ml.version();\n+ // not available until jar built\n+ Assert.assertEquals(MLContextUtil.VERSION_NOT_AVAILABLE, version);\n+ }\n+\n+ @Test\n+ public void testMLContextBuildTimeMessage() {\n+ System.out.println(\"MLContextTest - build time message\");\n+\n+ String buildTime = ml.buildTime();\n+ // not available until jar built\n+ Assert.assertEquals(MLContextUtil.BUILD_TIME_NOT_AVAILABLE, buildTime);\n+ }\n+\n+ @Test\n+ public void testMLContextCreateAndClose() {\n+ // MLContext created by the @BeforeClass method in MLContextTestBase\n+ // MLContext closed by the @AfterClass method in MLContextTestBase\n+ System.out.println(\"MLContextTest - create MLContext and close (without script execution)\");\n+ }\n+\n+ @Test\n+ public void testDataFrameToBinaryBlocks() {\n+ System.out.println(\"MLContextTest - DataFrame to binary blocks\");\n+\n+ List<String> list = new ArrayList<String>();\n+ list.add(\"1,2,3\");\n+ list.add(\"4,5,6\");\n+ list.add(\"7,8,9\");\n+ JavaRDD<String> javaRddString = sc.parallelize(list);\n+\n+ JavaRDD<Row> javaRddRow = javaRddString.map(new CommaSeparatedValueStringToDoubleArrayRow());\n+ List<StructField> fields = new ArrayList<StructField>();\n+ fields.add(DataTypes.createStructField(\"C1\", DataTypes.DoubleType, true));\n+ fields.add(DataTypes.createStructField(\"C2\", DataTypes.DoubleType, true));\n+ fields.add(DataTypes.createStructField(\"C3\", DataTypes.DoubleType, true));\n+ StructType schema = DataTypes.createStructType(fields);\n+ Dataset<Row> dataFrame = spark.createDataFrame(javaRddRow, schema);\n+\n+ JavaPairRDD<MatrixIndexes, MatrixBlock> binaryBlocks = MLContextConversionUtil\n+ .dataFrameToMatrixBinaryBlocks(dataFrame);\n+ Tuple2<MatrixIndexes, MatrixBlock> first = binaryBlocks.first();\n+ MatrixBlock mb = first._2();\n+ double[][] matrix = DataConverter.convertToDoubleMatrix(mb);\n+ Assert.assertArrayEquals(new double[] { 1.0, 2.0, 3.0 }, matrix[0], 0.0);\n+ Assert.assertArrayEquals(new double[] { 4.0, 5.0, 6.0 }, matrix[1], 0.0);\n+ Assert.assertArrayEquals(new double[] { 7.0, 8.0, 9.0 }, matrix[2], 0.0);\n+ }\n+\n+ @Test\n+ public void testGetTuple1DML() {\n+ System.out.println(\"MLContextTest - Get Tuple1<Matrix> DML\");\n+ JavaRDD<String> javaRddString = sc\n+ .parallelize(Stream.of(\"1,2,3\", \"4,5,6\", \"7,8,9\").collect(Collectors.toList()));\n+ JavaRDD<Row> javaRddRow = javaRddString.map(new CommaSeparatedValueStringToDoubleArrayRow());\n+ List<StructField> fields = new ArrayList<StructField>();\n+ fields.add(DataTypes.createStructField(\"C1\", DataTypes.DoubleType, true));\n+ fields.add(DataTypes.createStructField(\"C2\", DataTypes.DoubleType, true));\n+ fields.add(DataTypes.createStructField(\"C3\", DataTypes.DoubleType, true));\n+ StructType schema = DataTypes.createStructType(fields);\n+ Dataset<Row> df = spark.createDataFrame(javaRddRow, schema);\n+\n+ Script script = dml(\"N=M*2\").in(\"M\", df).out(\"N\");\n+ Tuple1<Matrix> tuple = ml.execute(script).getTuple(\"N\");\n+ double[][] n = tuple._1().to2DDoubleArray();\n+ Assert.assertEquals(2.0, n[0][0], 0);\n+ Assert.assertEquals(4.0, n[0][1], 0);\n+ Assert.assertEquals(6.0, n[0][2], 0);\n+ Assert.assertEquals(8.0, n[1][0], 0);\n+ Assert.assertEquals(10.0, n[1][1], 0);\n+ Assert.assertEquals(12.0, n[1][2], 0);\n+ Assert.assertEquals(14.0, n[2][0], 0);\n+ Assert.assertEquals(16.0, n[2][1], 0);\n+ Assert.assertEquals(18.0, n[2][2], 0);\n+ }\n+\n+ @Test\n+ public void testGetTuple2DML() {\n+ System.out.println(\"MLContextTest - Get Tuple2<Matrix,Double> DML\");\n+\n+ double[][] m = new double[][] { { 1, 2 }, { 3, 4 } };\n+\n+ Script script = dml(\"N=M*2;s=sum(N)\").in(\"M\", m).out(\"N\", \"s\");\n+ Tuple2<Matrix, Double> tuple = ml.execute(script).getTuple(\"N\", \"s\");\n+ double[][] n = tuple._1().to2DDoubleArray();\n+ double s = tuple._2();\n+ Assert.assertArrayEquals(new double[] { 2, 4 }, n[0], 0.0);\n+ Assert.assertArrayEquals(new double[] { 6, 8 }, n[1], 0.0);\n+ Assert.assertEquals(20.0, s, 0.0);\n+ }\n+\n+ @Test\n+ public void testGetTuple3DML() {\n+ System.out.println(\"MLContextTest - Get Tuple3<Long,Double,Boolean> DML\");\n+\n+ Script script = dml(\"a=1+2;b=a+0.5;c=TRUE;\").out(\"a\", \"b\", \"c\");\n+ Tuple3<Long, Double, Boolean> tuple = ml.execute(script).getTuple(\"a\", \"b\", \"c\");\n+ long a = tuple._1();\n+ double b = tuple._2();\n+ boolean c = tuple._3();\n+ Assert.assertEquals(3, a);\n+ Assert.assertEquals(3.5, b, 0.0);\n+ Assert.assertEquals(true, c);\n+ }\n+\n+ @Test\n+ public void testGetTuple4DML() {\n+ System.out.println(\"MLContextTest - Get Tuple4<Long,Double,Boolean,String> DML\");\n+\n+ Script script = dml(\"a=1+2;b=a+0.5;c=TRUE;d=\\\"yes it's \\\"+c\").out(\"a\", \"b\", \"c\", \"d\");\n+ Tuple4<Long, Double, Boolean, String> tuple = ml.execute(script).getTuple(\"a\", \"b\", \"c\", \"d\");\n+ long a = tuple._1();\n+ double b = tuple._2();\n+ boolean c = tuple._3();\n+ String d = tuple._4();\n+ Assert.assertEquals(3, a);\n+ Assert.assertEquals(3.5, b, 0.0);\n+ Assert.assertEquals(true, c);\n+ Assert.assertEquals(\"yes it's TRUE\", d);\n+ }\n+\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/setup.py", "new_path": "src/main/python/setup.py", "diff": "@@ -42,7 +42,7 @@ pillow_version = '2.0.0'\nREQUIRED_PACKAGES = [\n'numpy >= %s' % numpy_version,\n'scipy >= %s' % scipy_version,\n- 'pandas',\n+ 'pandas', 'scikit-learn',\n'Pillow >= %s' % pillow_version\n]\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/OptimizerUtils.java", "new_path": "src/main/java/org/apache/sysml/hops/OptimizerUtils.java", "diff": "@@ -1015,15 +1015,17 @@ public class OptimizerUtils\nreturn ( op==OpOp2.NOTEQUAL && val==0);\n}\n- public static double getBinaryOpSparsityConditionalSparseSafe( double sp1, OpOp2 op, LiteralOp lit )\n- {\n+ public static boolean isBinaryOpSparsityConditionalSparseSafe( OpOp2 op, LiteralOp lit ) {\ndouble val = HopRewriteUtils.getDoubleValueSafe(lit);\n-\nreturn ( (op==OpOp2.GREATER && val==0)\n||(op==OpOp2.LESS && val==0)\n||(op==OpOp2.NOTEQUAL && val==0)\n||(op==OpOp2.EQUAL && val!=0)\n- ||(op==OpOp2.MINUS && val==0)) ? sp1 : 1.0;\n+ ||(op==OpOp2.MINUS && val==0));\n+ }\n+\n+ public static double getBinaryOpSparsityConditionalSparseSafe( double sp1, OpOp2 op, LiteralOp lit ) {\n+ return isBinaryOpSparsityConditionalSparseSafe(op, lit) ? sp1 : 1.0;\n}\n/**\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateOuterProduct.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateOuterProduct.java", "diff": "@@ -78,7 +78,10 @@ public class TemplateOuterProduct extends TemplateBase {\npublic boolean merge(Hop hop, Hop input) {\nreturn !isClosed() &&\n(TemplateUtils.isBinaryMatrixRowVector(hop)\n- || HopRewriteUtils.isBinaryMatrixScalarOperation(hop));\n+ || HopRewriteUtils.isBinaryMatrixScalarOperation(hop)\n+ || (HopRewriteUtils.isBinary(hop, OpOp2.MULT)\n+ && HopRewriteUtils.isBinarySparseSafe(input)\n+ && !TemplateUtils.rContainsOuterProduct(input)));\n}\n@Override\n@@ -167,10 +170,10 @@ public class TemplateOuterProduct extends TemplateBase {\nCNode cdata2 = tmp.get(hop.getInput().get(1).getHopID());\nString primitiveOpName = ((BinaryOp)hop).getOp().toString();\n- if( HopRewriteUtils.isEqualSize(hop.getInput().get(0), hop.getInput().get(1)) ) {\n- Hop main = hop.getInput().get((cdata1 instanceof CNodeData) ? 0 : 1);\n- inHops2.put(\"_X\", main);\n- }\n+ if( TemplateUtils.isMatrix(hop.getInput().get(0)) && cdata1 instanceof CNodeData )\n+ inHops2.put(\"_X\", hop.getInput().get(0));\n+ if( TemplateUtils.isMatrix(hop.getInput().get(1)) && cdata2 instanceof CNodeData )\n+ inHops2.put(\"_X\", hop.getInput().get(1));\n//add lookups if required\ncdata1 = TemplateUtils.wrapLookupIfNecessary(cdata1, hop.getInput().get(0));\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateUtils.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateUtils.java", "diff": "@@ -430,4 +430,22 @@ public class TemplateUtils\nnode.setVisited();\nreturn ret;\n}\n+\n+ public static boolean containsOuterProduct(Hop hop) {\n+ hop.resetVisitStatus();\n+ boolean ret = rContainsOuterProduct(hop);\n+ hop.resetVisitStatus();\n+ return ret;\n+ }\n+\n+ public static boolean rContainsOuterProduct(Hop current) {\n+ if( current.isVisited() )\n+ return false;\n+ boolean ret = false;\n+ ret |= HopRewriteUtils.isOuterProductLikeMM(current);\n+ for( int i=0; i<current.getInput().size() && !ret; i++ )\n+ ret |= rContainsOuterProduct(current.getInput().get(i));\n+ current.setVisited();\n+ return ret;\n+ }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/rewrite/HopRewriteUtils.java", "new_path": "src/main/java/org/apache/sysml/hops/rewrite/HopRewriteUtils.java", "diff": "@@ -47,6 +47,7 @@ import org.apache.sysml.hops.IndexingOp;\nimport org.apache.sysml.hops.LeftIndexingOp;\nimport org.apache.sysml.hops.LiteralOp;\nimport org.apache.sysml.hops.MemoTable;\n+import org.apache.sysml.hops.OptimizerUtils;\nimport org.apache.sysml.hops.ParameterizedBuiltinOp;\nimport org.apache.sysml.hops.ReorgOp;\nimport org.apache.sysml.hops.TernaryOp;\n@@ -824,6 +825,18 @@ public class HopRewriteUtils\nreturn isBinary(hop, type) && hop.getParent().size() <= maxParents;\n}\n+ public static boolean isBinarySparseSafe(Hop hop) {\n+ if( !(hop instanceof BinaryOp) )\n+ return false;\n+ if( isBinary(hop, OpOp2.MULT) )\n+ return true;\n+ BinaryOp bop = (BinaryOp) hop;\n+ Hop lit = bop.getInput().get(0) instanceof LiteralOp ? bop.getInput().get(0) :\n+ bop.getInput().get(1) instanceof LiteralOp ? bop.getInput().get(1) : null;\n+ return lit != null && OptimizerUtils\n+ .isBinaryOpSparsityConditionalSparseSafe(bop.getOp(), (LiteralOp)lit);\n+ }\n+\npublic static boolean isBinaryMatrixScalarOperation(Hop hop) {\nreturn hop instanceof BinaryOp &&\n((hop.getInput().get(0).getDataType().isMatrix() && hop.getInput().get(1).getDataType().isScalar())\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/OuterProdTmplTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/OuterProdTmplTest.java", "diff": "@@ -43,6 +43,7 @@ public class OuterProdTmplTest extends AutomatedTestBase\nprivate static final String TEST_NAME6 = \"wdivmmbasic\";\nprivate static final String TEST_NAME7 = \"wdivmmTransposeOut\";\nprivate static final String TEST_NAME8 = \"wSparseUnsafeOuterProduct\";\n+ private static final String TEST_NAME9 = \"wdivmmNeq\";\nprivate static final String TEST_DIR = \"functions/codegen/\";\nprivate static final String TEST_CLASS_DIR = TEST_DIR + OuterProdTmplTest.class.getSimpleName() + \"/\";\n@@ -62,6 +63,7 @@ public class OuterProdTmplTest extends AutomatedTestBase\naddTestConfiguration( TEST_NAME6, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME6, new String[] { \"6\" }) );\naddTestConfiguration( TEST_NAME7, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME7, new String[] { \"7\" }) );\naddTestConfiguration( TEST_NAME8, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME8, new String[] { \"8\" }) );\n+ addTestConfiguration( TEST_NAME9, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME9, new String[] { \"9\" }) );\n}\n@Test\n@@ -169,6 +171,20 @@ public class OuterProdTmplTest extends AutomatedTestBase\ntestCodegenIntegrationWithInput( TEST_NAME8, true, ExecType.SPARK );\n}\n+ @Test\n+ public void testCodegenOuterProdRewrite9() {\n+ testCodegenIntegration( TEST_NAME9, true, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenOuterProd9() {\n+ testCodegenIntegration( TEST_NAME9, false, ExecType.CP );\n+ }\n+\n+ @Test\n+ public void testCodegenOuterProdRewrite9_sp() {\n+ testCodegenIntegrationWithInput( TEST_NAME9, true, ExecType.SPARK );\n+ }\nprivate void testCodegenIntegration( String testname, boolean rewrites, ExecType instType )\n{\n@@ -209,9 +225,12 @@ public class OuterProdTmplTest extends AutomatedTestBase\nif( testname.equals(TEST_NAME8) )\nAssert.assertTrue(!(heavyHittersContainsSubString(\"spoofOP\")\n|| heavyHittersContainsSubString(\"sp_spoofOP\")));\n- else if( !rewrites )\n+ else if( !rewrites ) {\nAssert.assertTrue(heavyHittersContainsSubString(\"spoofOP\")\n|| heavyHittersContainsSubString(\"sp_spoofOP\"));\n+ if( testname.equals(TEST_NAME9) )\n+ Assert.assertTrue(!heavyHittersContainsSubString(\"!=\"));\n+ }\n}\nfinally {\nrtplatform = platformOld;\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/wdivmmNeq.R", "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+args<-commandArgs(TRUE)\n+options(digits=22)\n+library(\"Matrix\")\n+\n+X = matrix( 3, 4000, 2000);\n+U = matrix( 4, 4000, 10);\n+V = matrix( 5, 2000, 10);\n+eps = 0.1;\n+S= ((X!=0)*(U%*%t(V)+eps)) %*% V;\n+writeMM(as(S, \"CsparseMatrix\"), paste(args[2], \"S\", sep=\"\"));\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/codegen/wdivmmNeq.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( 3, rows=4000, cols=2000);\n+U = matrix( 4, rows=4000, cols=10);\n+V = matrix( 5, rows=2000, cols=10);\n+while(FALSE){}\n+eps = 0.1;\n+S= ((X!=0)*(U%*%t(V)+eps)) %*% V;\n+\n+write(S,$1)\n" } ]

[ { "change_type": "MODIFY", "old_path": "dev/release/release-build.sh", "new_path": "dev/release/release-build.sh", "diff": "@@ -289,8 +289,8 @@ if [[ \"$RELEASE_PREPARE\" == \"true\" ]]; then\nfor i in *.zip *.tgz; do gpg --output $i.asc --detach-sig --armor $i; done\nrm -f *.md5\nfor i in *.zip *.tgz; do openssl md5 -hex $i | sed 's/MD5(\\([^)]*\\))= \\([0-9a-f]*\\)/\\2 *\\1/' > $i.md5; done\n- rm -f *.sha\n- for i in *.zip *.tgz; do shasum $i > $i.sha; done\n+ rm -f *.sha512\n+ for i in *.zip *.tgz; do shasum -a 512 $i > $i.sha512; done\ncd .. #exit $RELEASE_VERSION-$RELEASE_RC/\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/systemml/classloader.py", "new_path": "src/main/python/systemml/classloader.py", "diff": "__all__ = ['createJavaObject']\nimport os\n+import numpy as np\n+import pandas as pd\ntry:\nimport py4j.java_gateway\nfrom py4j.java_gateway import JavaObject\nfrom pyspark import SparkContext\n+ from pyspark.sql import SparkSession\nexcept ImportError:\nraise ImportError('Unable to import `pyspark`. Hint: Make sure you are running with PySpark.')\n+_initializedSparkSession = False\ndef _createJavaObject(sc, obj_type):\n+ # -----------------------------------------------------------------------------------\n+ # Avoids race condition between locking of metastore_db of Scala SparkSession and PySpark SparkSession.\n+ # This is done at toDF() rather than import level to avoid creation of SparkSession in worker processes.\n+ global _initializedSparkSession\n+ if not _initializedSparkSession:\n+ _initializedSparkSession = True\n+ SparkSession.builder.getOrCreate().createDataFrame(pd.DataFrame(np.array([[1,2],[3,4]])))\n+ # -----------------------------------------------------------------------------------\nif obj_type == 'mlcontext':\nreturn sc._jvm.org.apache.sysml.api.mlcontext.MLContext(sc._jsc)\nelif obj_type == 'dummy':\n" }, { "change_type": "MODIFY", "old_path": "src/main/python/systemml/mlcontext.py", "new_path": "src/main/python/systemml/mlcontext.py", "diff": "@@ -36,11 +36,7 @@ try:\nfrom pyspark import SparkContext\nfrom pyspark.conf import SparkConf\nimport pyspark.mllib.common\n- # -----------------------------------------------------------------------------------\n- # Avoids race condition between locking of metastore_db of Scala SparkSession and PySpark SparkSession\nfrom pyspark.sql import SparkSession\n- SparkSession.builder.getOrCreate().createDataFrame(pd.DataFrame(np.array([[1,2],[3,4]])))\n- # -----------------------------------------------------------------------------------\nexcept ImportError:\nraise ImportError('Unable to import `pyspark`. Hint: Make sure you are running with PySpark.')\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "<version>18</version>\n</parent>\n<groupId>org.apache.systemml</groupId>\n- <version>1.0.0-SNAPSHOT</version>\n+ <version>0.15.0-incubating</version>\n<artifactId>systemml</artifactId>\n<packaging>jar</packaging>\n<name>SystemML</name>\n<connection>scm:git:[email protected]:apache/systemml</connection>\n<developerConnection>scm:git:https://git-wip-us.apache.org/repos/asf/systemml</developerConnection>\n<url>https://git-wip-us.apache.org/repos/asf?p=systemml.git</url>\n- <tag>HEAD</tag>\n+ <tag>v0.15.0-incubating-rc1</tag>\n</scm>\n<issueManagement>\n<system>JIRA</system>\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "<version>18</version>\n</parent>\n<groupId>org.apache.systemml</groupId>\n- <version>1.0.0-incubating-SNAPSHOT</version>\n+ <version>0.15.0</version>\n<artifactId>systemml</artifactId>\n<packaging>jar</packaging>\n<name>SystemML</name>\n<connection>scm:git:[email protected]:apache/systemml</connection>\n<developerConnection>scm:git:https://git-wip-us.apache.org/repos/asf/systemml</developerConnection>\n<url>https://git-wip-us.apache.org/repos/asf?p=systemml.git</url>\n- <tag>HEAD</tag>\n+ <tag>v0.15.0-rc1</tag>\n</scm>\n<issueManagement>\n<system>JIRA</system>\n" } ]

[ { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysml/test/integration/mlcontext/algorithms/MLContextLinregTest.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.sysml.test.integration.mlcontext.algorithms;\n+\n+import static org.apache.sysml.api.mlcontext.ScriptFactory.dmlFromFile;\n+\n+import org.apache.log4j.Logger;\n+import org.apache.sysml.api.mlcontext.Script;\n+import org.apache.sysml.test.integration.mlcontext.MLContextTestBase;\n+import org.junit.Test;\n+\n+public class MLContextLinregTest extends MLContextTestBase {\n+ protected static Logger log = Logger.getLogger(MLContextLinregTest.class);\n+\n+ protected final static String TEST_SCRIPT_CG = \"scripts/algorithms/LinearRegCG.dml\";\n+ protected final static String TEST_SCRIPT_DS = \"scripts/algorithms/LinearRegDS.dml\";\n+ private final static double sparsity1 = 0.7; // dense\n+ private final static double sparsity2 = 0.1; // sparse\n+\n+ public enum LinregType {\n+ CG, DS,\n+ }\n+\n+ @Test\n+ public void testLinregCGSparse() {\n+ runLinregTestMLC(LinregType.CG, true);\n+ }\n+\n+ @Test\n+ public void testLinregCGDense() {\n+ runLinregTestMLC(LinregType.CG, false);\n+ }\n+\n+ @Test\n+ public void testLinregDSSparse() {\n+ runLinregTestMLC(LinregType.DS, true);\n+ }\n+\n+ @Test\n+ public void testLinregDSDense() {\n+ runLinregTestMLC(LinregType.DS, false);\n+ }\n+\n+ private void runLinregTestMLC(LinregType type, boolean sparse) {\n+\n+ double[][] X = getRandomMatrix(10, 3, 0, 1, sparse ? sparsity2 : sparsity1, 7);\n+ double[][] Y = getRandomMatrix(10, 1, 0, 10, 1.0, 3);\n+\n+ switch (type) {\n+ case CG:\n+ Script lrcg = dmlFromFile(TEST_SCRIPT_CG);\n+ lrcg.in(\"X\", X).in(\"y\", Y).in(\"$icpt\", \"0\").in(\"$tol\", \"0.000001\").in(\"$maxi\", \"0\").in(\"$reg\", \"0.000001\")\n+ .out(\"beta_out\");\n+ ml.execute(lrcg);\n+\n+ break;\n+\n+ case DS:\n+ Script lrds = dmlFromFile(TEST_SCRIPT_DS);\n+ lrds.in(\"X\", X).in(\"y\", Y).in(\"$icpt\", \"0\").in(\"$reg\", \"0.000001\").out(\"beta_out\");\n+ ml.execute(lrds);\n+\n+ break;\n+ }\n+ }\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/test_suites/java/org/apache/sysml/test/integration/mlcontext/algorithms/ZPackageSuite.java", "new_path": "src/test_suites/java/org/apache/sysml/test/integration/mlcontext/algorithms/ZPackageSuite.java", "diff": "@@ -26,7 +26,10 @@ import org.junit.runners.Suite;\n* Group together the tests in this package.\n*/\n@RunWith(Suite.class)\[email protected]({ MLContextUnivariateStatisticsTest.class })\[email protected]({\n+ MLContextUnivariateStatisticsTest.class,\n+ MLContextLinregTest.class\n+ })\n/** This class is just a holder for the above JUnit annotations. */\npublic class ZPackageSuite {\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "</dependency>\n<dependency>\n<groupId>org.jcuda</groupId>\n- <artifactId>jcufft</artifactId>\n+ <artifactId>jcusparse</artifactId>\n<version>${jcuda.version}</version>\n<scope>${jcuda.scope}</scope>\n</dependency>\n<dependency>\n<groupId>org.jcuda</groupId>\n- <artifactId>jcusparse</artifactId>\n+ <artifactId>jcusolver</artifactId>\n<version>${jcuda.version}</version>\n<scope>${jcuda.scope}</scope>\n</dependency>\n<dependency>\n<groupId>org.jcuda</groupId>\n- <artifactId>jcusolver</artifactId>\n+ <artifactId>jcudnn</artifactId>\n<version>${jcuda.version}</version>\n<scope>${jcuda.scope}</scope>\n</dependency>\n+\n<dependency>\n<groupId>org.jcuda</groupId>\n- <artifactId>jcurand</artifactId>\n+ <artifactId>jcuda-natives</artifactId>\n+ <classifier>${jcuda.os}-${jcuda.arch}</classifier>\n<version>${jcuda.version}</version>\n<scope>${jcuda.scope}</scope>\n</dependency>\n<dependency>\n<groupId>org.jcuda</groupId>\n- <artifactId>jnvgraph</artifactId>\n+ <artifactId>jcublas-natives</artifactId>\n+ <classifier>${jcuda.os}-${jcuda.arch}</classifier>\n<version>${jcuda.version}</version>\n<scope>${jcuda.scope}</scope>\n</dependency>\n<dependency>\n<groupId>org.jcuda</groupId>\n- <artifactId>jcudnn</artifactId>\n+ <artifactId>jcusparse-natives</artifactId>\n+ <classifier>${jcuda.os}-${jcuda.arch}</classifier>\n+ <version>${jcuda.version}</version>\n+ <scope>${jcuda.scope}</scope>\n+ </dependency>\n+ <dependency>\n+ <groupId>org.jcuda</groupId>\n+ <artifactId>jcusolver-natives</artifactId>\n+ <classifier>${jcuda.os}-${jcuda.arch}</classifier>\n+ <version>${jcuda.version}</version>\n+ <scope>${jcuda.scope}</scope>\n+ </dependency>\n+ <dependency>\n+ <groupId>org.jcuda</groupId>\n+ <artifactId>jcudnn-natives</artifactId>\n+ <classifier>${jcuda.os}-${jcuda.arch}</classifier>\n<version>${jcuda.version}</version>\n<scope>${jcuda.scope}</scope>\n</dependency>\n+ <!-- for all platforms, to be included in the extra jar -->\n<dependency>\n<groupId>org.jcuda</groupId>\n<artifactId>jcuda-natives</artifactId>\n- <classifier>${jcuda.os}-${jcuda.arch}</classifier>\n+ <classifier>windows-x86_64</classifier>\n<version>${jcuda.version}</version>\n<scope>${jcuda.scope}</scope>\n</dependency>\n<dependency>\n<groupId>org.jcuda</groupId>\n<artifactId>jcublas-natives</artifactId>\n- <classifier>${jcuda.os}-${jcuda.arch}</classifier>\n+ <classifier>windows-x86_64</classifier>\n<version>${jcuda.version}</version>\n<scope>${jcuda.scope}</scope>\n</dependency>\n<dependency>\n<groupId>org.jcuda</groupId>\n- <artifactId>jcufft-natives</artifactId>\n- <classifier>${jcuda.os}-${jcuda.arch}</classifier>\n+ <artifactId>jcusparse-natives</artifactId>\n+ <classifier>windows-x86_64</classifier>\n+ <version>${jcuda.version}</version>\n+ <scope>${jcuda.scope}</scope>\n+ </dependency>\n+ <dependency>\n+ <groupId>org.jcuda</groupId>\n+ <artifactId>jcusolver-natives</artifactId>\n+ <classifier>windows-x86_64</classifier>\n+ <version>${jcuda.version}</version>\n+ <scope>${jcuda.scope}</scope>\n+ </dependency>\n+ <dependency>\n+ <groupId>org.jcuda</groupId>\n+ <artifactId>jcudnn-natives</artifactId>\n+ <classifier>windows-x86_64</classifier>\n+ <version>${jcuda.version}</version>\n+ <scope>${jcuda.scope}</scope>\n+ </dependency>\n+\n+ <dependency>\n+ <groupId>org.jcuda</groupId>\n+ <artifactId>jcuda-natives</artifactId>\n+ <classifier>linux-x86_64</classifier>\n+ <version>${jcuda.version}</version>\n+ <scope>${jcuda.scope}</scope>\n+ </dependency>\n+ <dependency>\n+ <groupId>org.jcuda</groupId>\n+ <artifactId>jcublas-natives</artifactId>\n+ <classifier>linux-x86_64</classifier>\n<version>${jcuda.version}</version>\n<scope>${jcuda.scope}</scope>\n</dependency>\n<dependency>\n<groupId>org.jcuda</groupId>\n<artifactId>jcusparse-natives</artifactId>\n- <classifier>${jcuda.os}-${jcuda.arch}</classifier>\n+ <classifier>linux-x86_64</classifier>\n<version>${jcuda.version}</version>\n<scope>${jcuda.scope}</scope>\n</dependency>\n<dependency>\n<groupId>org.jcuda</groupId>\n<artifactId>jcusolver-natives</artifactId>\n- <classifier>${jcuda.os}-${jcuda.arch}</classifier>\n+ <classifier>linux-x86_64</classifier>\n<version>${jcuda.version}</version>\n<scope>${jcuda.scope}</scope>\n</dependency>\n<dependency>\n<groupId>org.jcuda</groupId>\n- <artifactId>jcurand-natives</artifactId>\n- <classifier>${jcuda.os}-${jcuda.arch}</classifier>\n+ <artifactId>jcudnn-natives</artifactId>\n+ <classifier>linux-x86_64</classifier>\n<version>${jcuda.version}</version>\n<scope>${jcuda.scope}</scope>\n</dependency>\n+\n<dependency>\n<groupId>org.jcuda</groupId>\n- <artifactId>jnvgraph-natives</artifactId>\n- <classifier>${jcuda.os}-${jcuda.arch}</classifier>\n+ <artifactId>jcuda-natives</artifactId>\n+ <classifier>linux-ppc_64</classifier>\n+ <version>${jcuda.version}</version>\n+ <scope>${jcuda.scope}</scope>\n+ </dependency>\n+ <dependency>\n+ <groupId>org.jcuda</groupId>\n+ <artifactId>jcublas-natives</artifactId>\n+ <classifier>linux-ppc_64</classifier>\n+ <version>${jcuda.version}</version>\n+ <scope>${jcuda.scope}</scope>\n+ </dependency>\n+ <dependency>\n+ <groupId>org.jcuda</groupId>\n+ <artifactId>jcusparse-natives</artifactId>\n+ <classifier>linux-ppc_64</classifier>\n+ <version>${jcuda.version}</version>\n+ <scope>${jcuda.scope}</scope>\n+ </dependency>\n+ <dependency>\n+ <groupId>org.jcuda</groupId>\n+ <artifactId>jcusolver-natives</artifactId>\n+ <classifier>linux-ppc_64</classifier>\n<version>${jcuda.version}</version>\n<scope>${jcuda.scope}</scope>\n</dependency>\n<dependency>\n<groupId>org.jcuda</groupId>\n<artifactId>jcudnn-natives</artifactId>\n- <classifier>${jcuda.os}-${jcuda.arch}</classifier>\n+ <classifier>linux-ppc_64</classifier>\n<version>${jcuda.version}</version>\n<scope>${jcuda.scope}</scope>\n</dependency>\n" }, { "change_type": "MODIFY", "old_path": "src/assembly/extra.xml", "new_path": "src/assembly/extra.xml", "diff": "</fileSet>\n</fileSets>\n- <!-- Include platform specific JCuda Jars -->\n+ <!-- Include JCuda Jars -->\n<dependencySets>\n<dependencySet>\n<includes>\n- <include>org.jcuda:*</include>\n+ <include>org.jcuda:jcuda:jar:${jcuda.version}</include>\n+ <include>org.jcuda:jcublas:jar:${jcuda.version}</include>\n+ <include>org.jcuda:jcusparse:jar:${jcuda.version}</include>\n+ <include>org.jcuda:jcusolver:jar:${jcuda.version}</include>\n+ <include>org.jcuda:jcudnn:jar:${jcuda.version}</include>\n+\n+ <!-- windows specific jcuda jars -->\n+ <include>org.jcuda:jcuda-natives:jar:windows-x86_64:${jcuda.version}</include>\n+ <include>org.jcuda:jcublas-natives:jar:windows-x86_64:${jcuda.version}</include>\n+ <include>org.jcuda:jcusparse-natives:jar:windows-x86_64:${jcuda.version}</include>\n+ <include>org.jcuda:jcusolver-natives:jar:windows-x86_64:${jcuda.version}</include>\n+ <include>org.jcuda:jcudnn-natives:jar:windows-x86_64:${jcuda.version}</include>\n+\n+ <!-- linux x86_64 specific jcuda jars -->\n+ <include>org.jcuda:jcuda-natives:jar:linux-x86_64:${jcuda.version}</include>\n+ <include>org.jcuda:jcublas-natives:jar:linux-x86_64:${jcuda.version}</include>\n+ <include>org.jcuda:jcusparse-natives:jar:linux-x86_64:${jcuda.version}</include>\n+ <include>org.jcuda:jcusolver-natives:jar:linux-x86_64:${jcuda.version}</include>\n+ <include>org.jcuda:jcudnn-natives:jar:linux-x86_64:${jcuda.version}</include>\n+\n+ <!-- linux ppc_64le specific jcuda jars -->\n+ <include>org.jcuda:jcuda-natives:jar:linux-ppc_64:${jcuda.version}</include>\n+ <include>org.jcuda:jcublas-natives:jar:linux-ppc_64:${jcuda.version}</include>\n+ <include>org.jcuda:jcusparse-natives:jar:linux-ppc_64:${jcuda.version}</include>\n+ <include>org.jcuda:jcusolver-natives:jar:linux-ppc_64:${jcuda.version}</include>\n+ <include>org.jcuda:jcudnn-natives:jar:linux-ppc_64:${jcuda.version}</include>\n</includes>\n<unpack>true</unpack>\n<scope>compile</scope>\n" } ]

[ { "change_type": "MODIFY", "old_path": "projects/breast_cancer/MachineLearning-Keras-ResNet50.ipynb", "new_path": "projects/breast_cancer/MachineLearning-Keras-ResNet50.ipynb", "diff": "\"from keras.applications.resnet50 import ResNet50\\n\",\n\"from keras.callbacks import ModelCheckpoint, TensorBoard\\n\",\n\"from keras.initializers import VarianceScaling\\n\",\n- \"from keras.layers import Dense, Dropout, Flatten, GlobalAveragePooling2D, Input, Lambda, merge\\n\",\n+ \"from keras.layers import Dense, Dropout, Flatten, GlobalAveragePooling2D, Input, Lambda\\n\",\n\"from keras.models import Model, load_model\\n\",\n\"from keras.optimizers import SGD\\n\",\n\"from keras.preprocessing.image import ImageDataGenerator\\n\",\n\"from keras.regularizers import l2\\n\",\n- \"from keras.utils import to_categorical\\n\",\n\"import matplotlib.pyplot as plt\\n\",\n\"import numpy as np\\n\",\n\"import pandas as pd\\n\",\n" }, { "change_type": "MODIFY", "old_path": "projects/breast_cancer/breastcancer/preprocessing.py", "new_path": "projects/breast_cancer/breastcancer/preprocessing.py", "diff": "@@ -71,6 +71,8 @@ def open_slide(slide_num, folder, training):\nslide = openslide.open_slide(filename)\nexcept OpenSlideError:\nslide = None\n+ except FileNotFoundError:\n+ slide = None\nreturn slide\n@@ -586,7 +588,7 @@ def preprocess(spark, slide_nums, folder=\"data\", training=True, tile_size=1024,\n# Append labels\nlabels_df = get_labels_df(folder)\nsamples_with_labels = (samples.map(\n- lambda tup: (tup[0], int(labels_df.at[tup[0],\"tumor_score\"]),\n+ lambda tup: (int(tup[0]), int(labels_df.at[tup[0],\"tumor_score\"]),\nfloat(labels_df.at[tup[0],\"molecular_score\"]), Vectors.dense(tup[1]))))\ndf = samples_with_labels.toDF([\"slide_num\", \"tumor_score\", \"molecular_score\", \"sample\"])\ndf = df.select(df.slide_num.astype(\"int\"), df.tumor_score.astype(\"int\"),\n" }, { "change_type": "MODIFY", "old_path": "projects/breast_cancer/preprocess.py", "new_path": "projects/breast_cancer/preprocess.py", "diff": "@@ -33,7 +33,7 @@ import pandas as pd\nfrom sklearn.model_selection import train_test_split\nfrom pyspark.sql import SparkSession\n-from breastcancer.preprocessing import add_row_indices, get_labels_df, preprocess, save\n+from breastcancer.preprocessing import add_row_indices, get_labels_df, preprocess, save, sample\n# Create new SparkSession\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/python/systemml/defmatrix.py", "new_path": "src/main/python/systemml/defmatrix.py", "diff": "@@ -321,7 +321,7 @@ def eval(outputs, execute=True):\ncheck_MLContext()\nreset()\noutputs = convert_outputs_to_list(outputs)\n- matrix.script.scriptString = perform_dfs(outputs, execute)\n+ matrix.script.setScriptString(perform_dfs(outputs, execute))\nif not execute:\nreset_output_flag(outputs)\nreturn matrix.script.scriptString\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/util/ConvolutionUtils.java", "new_path": "src/main/java/org/apache/sysml/runtime/util/ConvolutionUtils.java", "diff": "@@ -47,13 +47,31 @@ public class ConvolutionUtils {\nif(H <= 0 || R <= 0 || heightPadding < 0 || verticalStride < 0) {\nthrow new RuntimeException(\"Incorrect parameters: height=\" + H + \" filter_height=\" + R + \" stride=\" + verticalStride + \" pad=\" + heightPadding);\n}\n- return (H + 2 * heightPadding - R) / verticalStride + 1;\n+ long padded_image_height = H + 2 * heightPadding;\n+ long ret = (padded_image_height - R) / verticalStride + 1;\n+ if(ret <= 0 || ret > Integer.MAX_VALUE) {\n+ // Check for valid output activation height\n+ if(padded_image_height < R)\n+ throw new RuntimeException(\"Incorrect parameters: padded image height:\" + padded_image_height + \" cannot be less than filter_height:\" + R);\n+ else\n+ throw new RuntimeException(\"Incorrect parameters: height=\" + H + \" filter_height=\" + R + \" stride=\" + verticalStride + \" pad=\" + heightPadding + \" as P=\" + ret);\n+ }\n+ return ret;\n}\npublic static long getQ(long W, long S, long horizontalStride, long widthPadding) {\nif(W <= 0 || S <= 0 || widthPadding < 0 || horizontalStride < 0) {\nthrow new RuntimeException(\"Incorrect parameters: width=\" + W + \" filter_width=\" + S + \" stride=\" + horizontalStride + \" pad=\" + widthPadding);\n}\n- return (W + 2 * widthPadding - S) / horizontalStride + 1;\n+ long padded_image_width = W + 2 * widthPadding;\n+ long ret = (padded_image_width - S) / horizontalStride + 1;\n+ if(ret <= 0 || ret > Integer.MAX_VALUE) {\n+ // Check for valid output activation width\n+ if(padded_image_width < S)\n+ throw new RuntimeException(\"Incorrect parameters: padded image width:\" + padded_image_width + \" cannot be less than filter width:\" + S);\n+ else\n+ throw new RuntimeException(\"Incorrect parameters: width=\" + W + \" filter_width=\" + S + \" stride=\" + horizontalStride + \" pad=\" + widthPadding + \" as Q=\" + ret);\n+ }\n+ return ret;\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/AlgorithmGLM.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/codegen/AlgorithmGLM.java", "diff": "@@ -213,7 +213,7 @@ public class AlgorithmGLM extends AutomatedTestBase\naddArgs[0] = \"1\"; addArgs[1] = \"2.0\"; addArgs[2] = \"1\"; addArgs[3] = \"0.0\";\nbreak;\ncase BINOMIAL_PROBIT: //dfam, vpow, link, yneg\n- addArgs[0] = \"2\"; addArgs[1] = \"0.0\"; addArgs[2] = \"3\"; addArgs[3] = \"2\";\n+ addArgs[0] = \"2\"; addArgs[1] = \"0.0\"; addArgs[2] = \"3\"; addArgs[3] = \"0\";\nparam4Name = \"yneg=\";\nbreak;\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/DMLScript.java", "new_path": "src/main/java/org/apache/sysml/api/DMLScript.java", "diff": "@@ -77,7 +77,6 @@ import org.apache.sysml.parser.ParserWrapper;\nimport org.apache.sysml.runtime.DMLRuntimeException;\nimport org.apache.sysml.runtime.DMLScriptException;\nimport org.apache.sysml.runtime.controlprogram.Program;\n-import org.apache.sysml.runtime.controlprogram.caching.CacheStatistics;\nimport org.apache.sysml.runtime.controlprogram.caching.CacheableData;\nimport org.apache.sysml.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysml.runtime.controlprogram.context.ExecutionContextFactory;\n@@ -829,11 +828,9 @@ public class DMLScript\n//reset statistics (required if multiple scripts executed in one JVM)\nStatistics.resetNoOfExecutedJobs();\n- if( STATISTICS ) {\n- CacheStatistics.reset();\n+ if( STATISTICS )\nStatistics.reset();\n}\n- }\nprivate static void checkSecuritySetup(DMLConfig config)\nthrows IOException, DMLRuntimeException\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/mlcontext/ScriptExecutor.java", "new_path": "src/main/java/org/apache/sysml/api/mlcontext/ScriptExecutor.java", "diff": "@@ -48,7 +48,6 @@ import org.apache.sysml.parser.ParserWrapper;\nimport org.apache.sysml.runtime.DMLRuntimeException;\nimport org.apache.sysml.runtime.controlprogram.LocalVariableMap;\nimport org.apache.sysml.runtime.controlprogram.Program;\n-import org.apache.sysml.runtime.controlprogram.caching.CacheStatistics;\nimport org.apache.sysml.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysml.runtime.controlprogram.context.ExecutionContextFactory;\nimport org.apache.sysml.runtime.instructions.gpu.context.GPUContextPool;\n@@ -376,11 +375,9 @@ public class ScriptExecutor {\nsetGlobalFlags();\n// reset all relevant summary statistics\nStatistics.resetNoOfExecutedJobs();\n- if (statistics) {\n- CacheStatistics.reset();\n+ if (statistics)\nStatistics.reset();\n}\n- }\n/**\n* Perform any necessary cleanup operations after program execution.\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/parfor/RemoteDPParWorkerReducer.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/parfor/RemoteDPParWorkerReducer.java", "diff": "@@ -33,7 +33,6 @@ import org.apache.sysml.conf.ConfigurationManager;\nimport org.apache.sysml.runtime.DMLRuntimeException;\nimport org.apache.sysml.runtime.controlprogram.ParForProgramBlock.PDataPartitionFormat;\nimport org.apache.sysml.runtime.controlprogram.ParForProgramBlock.PartitionFormat;\n-import org.apache.sysml.runtime.controlprogram.caching.CacheStatistics;\nimport org.apache.sysml.runtime.controlprogram.caching.CacheableData;\nimport org.apache.sysml.runtime.controlprogram.caching.MatrixObject;\nimport org.apache.sysml.runtime.controlprogram.parfor.Task.TaskType;\n@@ -189,11 +188,8 @@ public class RemoteDPParWorkerReducer extends ParWorker\n//always reset stats because counters per map task (for case of JVM reuse)\nif( DMLScript.STATISTICS && !InfrastructureAnalyzer.isLocalMode(job) )\n- {\n- CacheStatistics.reset();\nStatistics.reset();\n}\n- }\n@Override\npublic void close()\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/parfor/RemoteParWorkerMapper.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/parfor/RemoteParWorkerMapper.java", "diff": "@@ -33,7 +33,6 @@ import org.apache.sysml.api.DMLScript;\nimport org.apache.sysml.conf.ConfigurationManager;\nimport org.apache.sysml.runtime.controlprogram.LocalVariableMap;\nimport org.apache.sysml.runtime.controlprogram.ParForProgramBlock;\n-import org.apache.sysml.runtime.controlprogram.caching.CacheStatistics;\nimport org.apache.sysml.runtime.controlprogram.caching.CacheableData;\nimport org.apache.sysml.runtime.controlprogram.caching.MatrixObject;\nimport org.apache.sysml.runtime.controlprogram.parfor.stat.InfrastructureAnalyzer;\n@@ -212,11 +211,8 @@ public class RemoteParWorkerMapper extends ParWorker //MapReduceBase not requir\n//always reset stats because counters per map task (for case of JVM reuse)\nif( DMLScript.STATISTICS && !InfrastructureAnalyzer.isLocalMode(job) )\n- {\n- CacheStatistics.reset();\nStatistics.reset();\n}\n- }\n@Override\npublic void close()\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/utils/Statistics.java", "new_path": "src/main/java/org/apache/sysml/utils/Statistics.java", "diff": "@@ -407,6 +407,8 @@ public class Statistics\ncodegenEnumEvalP.reset();\ncodegenCompileTime.reset();\ncodegenClassCompileTime.reset();\n+ codegenPlanCacheHits.reset();\n+ codegenPlanCacheTotal.reset();\nparforOptCount = 0;\nparforOptTime = 0;\n@@ -417,6 +419,8 @@ public class Statistics\nlTotalLixUIP.reset();\nlTotalUIPVar.reset();\n+ CacheStatistics.reset();\n+\nresetJITCompileTime();\nresetJVMgcTime();\nresetJVMgcCount();\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/perftest/genMultinomialData.sh", "new_path": "scripts/perftest/genMultinomialData.sh", "diff": "@@ -33,31 +33,31 @@ SPARSE_SP=0.01\nexport HADOOP_CLIENT_OPTS=\"-Xmx2048m -Xms2048m -Xmn256m\"\n#generate XS scenarios (80MB)\n-${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 10000 1000 $DENSE_SP 150 0 $BASE/X10k_1k_dense_k150 $BASE/y10k_1k_dense_k150 $FORMAT 1\n-${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 10000 1000 $SPARSE_SP 150 0 $BASE/X10k_1k_sparse_k150 $BASE/y10k_1k_sparse_k150 $FORMAT 1\n-${CMD} -f extractTestData.dml $DASH-args $BASE/X10k_1k_dense_k150 $BASE/y10k_1k_dense_k150 $BASE/X10k_1k_dense_k150_test $BASE/y10k_1k_dense_k150_test $FORMAT\n-${CMD} -f extractTestData.dml $DASH-args $BASE/X10k_1k_sparse_k150 $BASE/y10k_1k_sparse_k150 $BASE/X10k_1k_sparse_k150_test $BASE/y10k_1k_sparse_k150_test $FORMAT\n+${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 10000 1000 $DENSE_SP 5 0 $BASE/X10k_1k_dense_k5 $BASE/y10k_1k_dense_k5 $FORMAT 1\n+${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 10000 1000 $SPARSE_SP 5 0 $BASE/X10k_1k_sparse_k5 $BASE/y10k_1k_sparse_k5 $FORMAT 1\n+${CMD} -f extractTestData.dml $DASH-args $BASE/X10k_1k_dense_k5 $BASE/y10k_1k_dense_k5 $BASE/X10k_1k_dense_k5_test $BASE/y10k_1k_dense_k5_test $FORMAT\n+${CMD} -f extractTestData.dml $DASH-args $BASE/X10k_1k_sparse_k5 $BASE/y10k_1k_sparse_k5 $BASE/X10k_1k_sparse_k5_test $BASE/y10k_1k_sparse_k5_test $FORMAT\n##generate S scenarios (80MB)\n-#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 100000 1000 $DENSE_SP 150 0 $BASE/X100k_1k_dense_k150 $BASE/y100k_1k_dense_k150 $FORMAT 1\n-#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 100000 1000 $SPARSE_SP 150 0 $BASE/X100k_1k_sparse_k150 $BASE/y100k_1k_sparse_k150 $FORMAT 1\n-#${CMD} -f extractTestData.dml $DASH-args $BASE/X100k_1k_dense_k150 $BASE/y100k_1k_dense_k150 $BASE/X100k_1k_dense_k150_test $BASE/y100k_1k_dense_k150_test $FORMAT\n-#${CMD} -f extractTestData.dml $DASH-args $BASE/X100k_1k_sparse_k150 $BASE/y100k_1k_sparse_k150 $BASE/X100k_1k_sparse_k150_test $BASE/y100k_1k_sparse_k150_test $FORMAT\n+#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 100000 1000 $DENSE_SP 5 0 $BASE/X100k_1k_dense_k5 $BASE/y100k_1k_dense_k5 $FORMAT 1\n+#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 100000 1000 $SPARSE_SP 5 0 $BASE/X100k_1k_sparse_k5 $BASE/y100k_1k_sparse_k5 $FORMAT 1\n+#${CMD} -f extractTestData.dml $DASH-args $BASE/X100k_1k_dense_k5 $BASE/y100k_1k_dense_k5 $BASE/X100k_1k_dense_k5_test $BASE/y100k_1k_dense_k5_test $FORMAT\n+#${CMD} -f extractTestData.dml $DASH-args $BASE/X100k_1k_sparse_k5 $BASE/y100k_1k_sparse_k5 $BASE/X100k_1k_sparse_k5_test $BASE/y100k_1k_sparse_k5_test $FORMAT\n#\n##generate M scenarios (8GB)\n-#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 1000000 1000 $DENSE_SP 150 0 $BASE/X1M_1k_dense_k150 $BASE/y1M_1k_dense_k150 $FORMAT 1\n-#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 1000000 1000 $SPARSE_SP 150 0 $BASE/X1M_1k_sparse_k150 $BASE/y1M_1k_sparse_k150 $FORMAT 1\n-#${CMD} -f extractTestData.dml $DASH-args $BASE/X1M_1k_dense_k150 $BASE/y1M_1k_dense_k150 $BASE/X1M_1k_dense_k150_test $BASE/y1M_1k_dense_k150_test $FORMAT\n-#${CMD} -f extractTestData.dml $DASH-args $BASE/X1M_1k_sparse_k150 $BASE/y1M_1k_sparse_k150 $BASE/X1M_1k_sparse_k150_test $BASE/y1M_1k_sparse_k150_test $FORMAT\n+#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 1000000 1000 $DENSE_SP 5 0 $BASE/X1M_1k_dense_k5 $BASE/y1M_1k_dense_k5 $FORMAT 1\n+#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 1000000 1000 $SPARSE_SP 5 0 $BASE/X1M_1k_sparse_k5 $BASE/y1M_1k_sparse_k5 $FORMAT 1\n+#${CMD} -f extractTestData.dml $DASH-args $BASE/X1M_1k_dense_k5 $BASE/y1M_1k_dense_k5 $BASE/X1M_1k_dense_k5_test $BASE/y1M_1k_dense_k5_test $FORMAT\n+#${CMD} -f extractTestData.dml $DASH-args $BASE/X1M_1k_sparse_k5 $BASE/y1M_1k_sparse_k5 $BASE/X1M_1k_sparse_k5_test $BASE/y1M_1k_sparse_k5_test $FORMAT\n#\n##generate L scenarios (80GB)\n-#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 10000000 1000 $DENSE_SP 150 0 $BASE/X10M_1k_dense_k150 $BASE/y10M_1k_dense_k150 $FORMAT 1\n-#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 10000000 1000 $SPARSE_SP 150 0 $BASE/X10M_1k_sparse_k150 $BASE/y10M_1k_sparse_k150 $FORMAT 1\n-#${CMD} -f extractTestData.dml $DASH-args $BASE/X10M_1k_dense_k150 $BASE/y10M_1k_dense_k150 $BASE/X10M_1k_dense_k150_test $BASE/y10M_1k_dense_k150_test $FORMAT\n-#${CMD} -f extractTestData.dml $DASH-args $BASE/X10M_1k_sparse_k150 $BASE/y10M_1k_sparse_k150 $BASE/X10M_1k_sparse_k150_test $BASE/y10M_1k_sparse_k150_test $FORMAT\n+#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 10000000 1000 $DENSE_SP 5 0 $BASE/X10M_1k_dense_k5 $BASE/y10M_1k_dense_k5 $FORMAT 1\n+#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 10000000 1000 $SPARSE_SP 5 0 $BASE/X10M_1k_sparse_k5 $BASE/y10M_1k_sparse_k5 $FORMAT 1\n+#${CMD} -f extractTestData.dml $DASH-args $BASE/X10M_1k_dense_k5 $BASE/y10M_1k_dense_k5 $BASE/X10M_1k_dense_k5_test $BASE/y10M_1k_dense_k5_test $FORMAT\n+#${CMD} -f extractTestData.dml $DASH-args $BASE/X10M_1k_sparse_k5 $BASE/y10M_1k_sparse_k5 $BASE/X10M_1k_sparse_k5_test $BASE/y10M_1k_sparse_k5_test $FORMAT\n#\n##generate LARGE scenarios (800GB)\n-#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 100000000 1000 $DENSE_SP 150 0 $BASE/X100M_1k_dense_k150 $BASE/y100M_1k_dense_k150 $FORMAT 1\n-#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 100000000 1000 $SPARSE_SP 150 0 $BASE/X100M_1k_sparse_k150 $BASE/y100M_1k_sparse_k150 $FORMAT 1\n-#${CMD} -f extractTestData.dml $DASH-args $BASE/X100M_1k_dense_k150 $BASE/y100M_1k_dense_k150 $BASE/X100M_1k_dense_k150_test $BASE/y100M_1k_dense_k150_test $FORMAT\n-#${CMD} -f extractTestData.dml $DASH-args $BASE/X100M_1k_sparse_k150 $BASE/y100M_1k_sparse_k150 $BASE/X100M_1k_sparse_k150_test $BASE/y100M_1k_sparse_k150_test $FORMAT\n+#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 100000000 1000 $DENSE_SP 5 0 $BASE/X100M_1k_dense_k5 $BASE/y100M_1k_dense_k5 $FORMAT 1\n+#${CMD} -f ../datagen/genRandData4Multinomial.dml $DASH-args 100000000 1000 $SPARSE_SP 5 0 $BASE/X100M_1k_sparse_k5 $BASE/y100M_1k_sparse_k5 $FORMAT 1\n+#${CMD} -f extractTestData.dml $DASH-args $BASE/X100M_1k_dense_k5 $BASE/y100M_1k_dense_k5 $BASE/X100M_1k_dense_k5_test $BASE/y100M_1k_dense_k5_test $FORMAT\n+#${CMD} -f extractTestData.dml $DASH-args $BASE/X100M_1k_sparse_k5 $BASE/y100M_1k_sparse_k5 $BASE/X100M_1k_sparse_k5_test $BASE/y100M_1k_sparse_k5_test $FORMAT\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/runAllMultinomial.sh", "new_path": "scripts/perftest/runAllMultinomial.sh", "diff": "@@ -47,18 +47,18 @@ do\nfor f in \"runNaiveBayes\"\ndo\necho \"-- Running \"$f\" on \"$d\" (all configs)\" >> times.txt;\n- ./${f}.sh ${BASE}/X${d}_k150 ${BASE}/y${d}_k150 150 ${BASE} $2 &> logs/${f}_${d}_k150.out;\n+ ./${f}.sh ${BASE}/X${d}_k5 ${BASE}/y${d}_k5 5 ${BASE} $2 &> logs/${f}_${d}_k5.out;\ndone\n# run with the parameter setting maximum of iterations\nfor f in \"runMultiLogReg\" \"runMSVM\"\ndo\necho \"-- Running \"$f\" on \"$d\" (all configs)\" >> times.txt;\n- ./${f}.sh ${BASE}/X${d}_k150 ${BASE}/y${d}_k150 150 ${BASE} $2 ${MAXITR} &> logs/${f}_${d}_k150.out;\n+ ./${f}.sh ${BASE}/X${d}_k5 ${BASE}/y${d}_k5 5 ${BASE} $2 ${MAXITR} &> logs/${f}_${d}_k5.out;\ndone\ndone\n#run KDD only on naive bayes (see binomial for the others)\n-#./runNaiveBayes.sh ${BASE0}/X_KDD_k150 ${BASE}/y_KDD_k150 150 &> logs/runNaiveBayes__KDD_k150.out;\n-#./runNaiveBayes.sh ${BASE0}/X_KDD ${BASE}/y_KDD 150 &> logs/runNaiveBayes__KDD_k150.out;\n+#./runNaiveBayes.sh ${BASE0}/X_KDD_k5 ${BASE}/y_KDD_k5 5 &> logs/runNaiveBayes__KDD_k5.out;\n+#./runNaiveBayes.sh ${BASE0}/X_KDD ${BASE}/y_KDD 5 &> logs/runNaiveBayes__KDD_k5.out;\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/perftest/genClusteringData.sh", "new_path": "scripts/perftest/genClusteringData.sh", "diff": "@@ -31,22 +31,22 @@ SPARSE_SP=0.01\nexport HADOOP_CLIENT_OPTS=\"-Xmx2048m -Xms2048m -Xmn256m\"\n#generate XS scenarios (80MB)\n-${CMD} -f ../datagen/genRandData4Kmeans.dml $DASH-nvargs nr=10000 nf=1000 nc=50 dc=10.0 dr=1.0 fbf=100.0 cbf=100.0 X=$BASE/X10k_1k_dense C=$BASE/C10k_1k_dense Y=$BASE/y10k_1k_dense YbyC=$BASE/YbyC10k_1k_dense fmt=$FORMAT\n+${CMD} -f ../datagen/genRandData4Kmeans.dml $DASH-nvargs nr=10000 nf=1000 nc=5 dc=10.0 dr=1.0 fbf=100.0 cbf=100.0 X=$BASE/X10k_1k_dense C=$BASE/C10k_1k_dense Y=$BASE/y10k_1k_dense YbyC=$BASE/YbyC10k_1k_dense fmt=$FORMAT\n${CMD} -f extractTestData.dml $DASH-args $BASE/X10k_1k_dense $BASE/y10k_1k_dense $BASE/X10k_1k_dense_test $BASE/y10k_1k_dense_test $FORMAT\n#generate S scenarios (800MB)\n-#${CMD} -f ../datagen/genRandData4Kmeans.dml $DASH-nvargs nr=100000 nf=1000 nc=50 dc=10.0 dr=1.0 fbf=100.0 cbf=100.0 X=$BASE/X100k_1k_dense C=$BASE/C100k_1k_dense Y=$BASE/y100k_1k_dense YbyC=$BASE/YbyC100k_1k_dense fmt=$FORMAT\n+#${CMD} -f ../datagen/genRandData4Kmeans.dml $DASH-nvargs nr=100000 nf=1000 nc=5 dc=10.0 dr=1.0 fbf=100.0 cbf=100.0 X=$BASE/X100k_1k_dense C=$BASE/C100k_1k_dense Y=$BASE/y100k_1k_dense YbyC=$BASE/YbyC100k_1k_dense fmt=$FORMAT\n#${CMD} -f extractTestData.dml $DASH-args $BASE/X100k_1k_dense $BASE/y100k_1k_dense $BASE/X100k_1k_dense_test $BASE/y100k_1k_dense_test $FORMAT\n#generate M scenarios (8GB)\n-#${CMD} -f ../datagen/genRandData4Kmeans.dml $DASH-nvargs nr=1000000 nf=1000 nc=50 dc=10.0 dr=1.0 fbf=100.0 cbf=100.0 X=$BASE/X1M_1k_dense C=$BASE/C1M_1k_dense Y=$BASE/y1M_1k_dense YbyC=$BASE/YbyC1M_1k_dense fmt=$FORMAT\n+#${CMD} -f ../datagen/genRandData4Kmeans.dml $DASH-nvargs nr=1000000 nf=1000 nc=5 dc=10.0 dr=1.0 fbf=100.0 cbf=100.0 X=$BASE/X1M_1k_dense C=$BASE/C1M_1k_dense Y=$BASE/y1M_1k_dense YbyC=$BASE/YbyC1M_1k_dense fmt=$FORMAT\n#${CMD} -f extractTestData.dml $DASH-args $BASE/X1M_1k_dense $BASE/y1M_1k_dense $BASE/X1M_1k_dense_test $BASE/y1M_1k_dense_test $FORMAT\n#generate L scenarios (80GB)\n-#${CMD} -f ../datagen/genRandData4Kmeans.dml $DASH-nvargs nr=10000000 nf=1000 nc=50 dc=10.0 dr=1.0 fbf=100.0 cbf=100.0 X=$BASE/X10M_1k_dense C=$BASE/C10M_1k_dense Y=$BASE/y10M_1k_dense YbyC=$BASE/YbyC10M_1k_dense fmt=$FORMAT\n+#${CMD} -f ../datagen/genRandData4Kmeans.dml $DASH-nvargs nr=10000000 nf=1000 nc=5 dc=10.0 dr=1.0 fbf=100.0 cbf=100.0 X=$BASE/X10M_1k_dense C=$BASE/C10M_1k_dense Y=$BASE/y10M_1k_dense YbyC=$BASE/YbyC10M_1k_dense fmt=$FORMAT\n#${CMD} -f extractTestData.dml $DASH-args $BASE/X10M_1k_dense $BASE/y10M_1k_dense $BASE/X10M_1k_dense_test $BASE/y10M_1k_dense_test $FORMAT\n#generate LARGE scenarios (800GB)\n-#${CMD} -f ../datagen/genRandData4Kmeans.dml $DASH-nvargs nr=100000000 nf=1000 nc=50 dc=10.0 dr=1.0 fbf=100.0 cbf=100.0 X=$BASE/X100M_1k_dense C=$BASE/C100M_1k_dense Y=$BASE/y100M_1k_dense YbyC=$BASE/YbyC100M_1k_dense fmt=$FORMAT\n+#${CMD} -f ../datagen/genRandData4Kmeans.dml $DASH-nvargs nr=100000000 nf=1000 nc=5 dc=10.0 dr=1.0 fbf=100.0 cbf=100.0 X=$BASE/X100M_1k_dense C=$BASE/C100M_1k_dense Y=$BASE/y100M_1k_dense YbyC=$BASE/YbyC100M_1k_dense fmt=$FORMAT\n#${CMD} -f extractTestData.dml $DASH-args $BASE/X100M_1k_dense $BASE/y100M_1k_dense $BASE/X100M_1k_dense_test $BASE/y100M_1k_dense_test $FORMAT\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/runKmeans.sh", "new_path": "scripts/perftest/runKmeans.sh", "diff": "@@ -29,7 +29,7 @@ export HADOOP_CLIENT_OPTS=\"-Xmx2048m -Xms2048m -Xmn256m\"\n#training\ntstart=$SECONDS\n-${CMD} -f ../algorithms/Kmeans.dml $DASH-explain $DASH-stats $DASH-nvargs X=$1 k=50 C=${BASE}/centroids.mtx maxi=$2 tol=0.0001\n+${CMD} -f ../algorithms/Kmeans.dml $DASH-explain $DASH-stats $DASH-nvargs X=$1 k=5 C=${BASE}/centroids.mtx maxi=$2 tol=0.0001\nttrain=$(($SECONDS - $tstart - 3))\necho \"Kmeans train on \"$1\": \"$ttrain >> times.txt\n" } ]

[ { "change_type": "MODIFY", "old_path": "conf/SystemML-config.xml.template", "new_path": "conf/SystemML-config.xml.template", "diff": "<!-- enables operator fusion via code generation, experimental feature -->\n<codegen.enabled>false</codegen.enabled>\n+ <!-- set the codegen java compiler (auto, janino, javac) -->\n+ <codegen.compiler>auto</codegen.compiler>\n+\n<!-- if codegen.enabled, enables source code caching of fused operators -->\n<codegen.plancache>false</codegen.plancache>\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/conf/DMLConfig.java", "new_path": "src/main/java/org/apache/sysml/conf/DMLConfig.java", "diff": "@@ -40,6 +40,7 @@ import org.apache.commons.logging.LogFactory;\nimport org.apache.hadoop.fs.FileSystem;\nimport org.apache.hadoop.fs.Path;\nimport org.apache.sysml.hops.OptimizerUtils;\n+import org.apache.sysml.hops.codegen.SpoofCompiler.CompilerType;\nimport org.apache.sysml.parser.ParseException;\nimport org.apache.sysml.runtime.DMLRuntimeException;\nimport org.apache.sysml.runtime.io.IOUtilFunctions;\n@@ -74,6 +75,7 @@ public class DMLConfig\npublic static final String COMPRESSED_LINALG = \"compressed.linalg\";\npublic static final String NATIVE_BLAS = \"native.blas\";\npublic static final String CODEGEN = \"codegen.enabled\"; //boolean\n+ public static final String CODEGEN_COMPILER = \"codegen.compiler\"; //see SpoofCompiler.CompilerType\npublic static final String CODEGEN_PLANCACHE = \"codegen.plancache\"; //boolean\npublic static final String CODEGEN_LITERALS = \"codegen.literals\"; //1..heuristic, 2..always\npublic static final String EXTRA_FINEGRAINED_STATS = \"systemml.stats.finegrained\"; //boolean\n@@ -119,6 +121,7 @@ public class DMLConfig\n_defaultVals.put(CP_PARALLEL_IO, \"true\" );\n_defaultVals.put(COMPRESSED_LINALG, \"false\" );\n_defaultVals.put(CODEGEN, \"false\" );\n+ _defaultVals.put(CODEGEN_COMPILER, CompilerType.AUTO.name() );\n_defaultVals.put(CODEGEN_PLANCACHE, \"true\" );\n_defaultVals.put(CODEGEN_LITERALS, \"1\" );\n_defaultVals.put(NATIVE_BLAS, \"none\" );\n@@ -126,7 +129,6 @@ public class DMLConfig\n_defaultVals.put(STATS_MAX_WRAP_LEN, \"30\" );\n_defaultVals.put(EXTRA_GPU_STATS, \"false\" );\n_defaultVals.put(EXTRA_DNN_STATS, \"false\" );\n-\n_defaultVals.put(GPU_MEMORY_UTILIZATION_FACTOR, \"0.9\" );\n_defaultVals.put(AVAILABLE_GPUS, \"-1\");\n}\n@@ -408,7 +410,8 @@ public class DMLConfig\nNUM_REDUCERS, DEFAULT_BLOCK_SIZE,\nYARN_APPMASTER, YARN_APPMASTERMEM, YARN_MAPREDUCEMEM,\nCP_PARALLEL_OPS, CP_PARALLEL_IO, NATIVE_BLAS,\n- COMPRESSED_LINALG, CODEGEN, CODEGEN_LITERALS, CODEGEN_PLANCACHE,\n+ COMPRESSED_LINALG,\n+ CODEGEN, CODEGEN_COMPILER, CODEGEN_PLANCACHE, CODEGEN_LITERALS,\nEXTRA_GPU_STATS, EXTRA_DNN_STATS, EXTRA_FINEGRAINED_STATS, STATS_MAX_WRAP_LEN,\nAVAILABLE_GPUS\n};\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/SpoofCompiler.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/SpoofCompiler.java", "diff": "@@ -36,6 +36,8 @@ import org.apache.log4j.Logger;\nimport org.apache.sysml.api.DMLException;\nimport org.apache.sysml.api.DMLScript;\nimport org.apache.sysml.api.DMLScript.RUNTIME_PLATFORM;\n+import org.apache.sysml.conf.ConfigurationManager;\n+import org.apache.sysml.conf.DMLConfig;\nimport org.apache.sysml.hops.codegen.cplan.CNode;\nimport org.apache.sysml.hops.codegen.cplan.CNodeCell;\nimport org.apache.sysml.hops.codegen.cplan.CNodeData;\n@@ -114,6 +116,7 @@ public class SpoofCompiler\npublic static final PlanSelector PLAN_SEL_POLICY = PlanSelector.FUSE_COST_BASED_V2;\npublic enum CompilerType {\n+ AUTO,\nJAVAC,\nJANINO,\n}\n@@ -484,7 +487,11 @@ public class SpoofCompiler\n}\npublic static void setExecTypeSpecificJavaCompiler() {\n- JAVA_COMPILER = OptimizerUtils.isSparkExecutionMode() ?\n+ DMLConfig conf = ConfigurationManager.getDMLConfig();\n+ String compiler = conf.getTextValue(DMLConfig.CODEGEN_COMPILER);\n+ CompilerType type = CompilerType.valueOf(compiler.toUpperCase());\n+ JAVA_COMPILER = (type != CompilerType.AUTO) ? type :\n+ OptimizerUtils.isSparkExecutionMode() ?\nCompilerType.JANINO : CompilerType.JAVAC;\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/ParForProgramBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/ParForProgramBlock.java", "diff": "@@ -1279,7 +1279,8 @@ public class ParForProgramBlock extends ForProgramBlock\n}\nbreak;\ncase MATRIX:\n- //currently we do not create any unscoped matrix object outputs\n+ case FRAME:\n+ //currently we do not create any unscoped matrix or frame outputs\n//because metadata (e.g., outputinfo) not known at this place.\nbreak;\ncase UNKNOWN:\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/parfor/opt/OptimizerRuleBased.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/parfor/opt/OptimizerRuleBased.java", "diff": "@@ -456,8 +456,7 @@ public class OptimizerRuleBased extends Optimizer\n{\nboolean ret = false;\n- if( !n.isLeaf() )\n- {\n+ if( !n.isLeaf() ) {\nfor( OptNode cn : n.getChilds() )\nif( cn.getNodeType() != NodeType.FUNCCALL ) //prevent conflicts with aliases\nret |= rFindDataPartitioningCandidates( cn, cand, vars, thetaM );\n@@ -467,7 +466,7 @@ public class OptimizerRuleBased extends Optimizer\n{\nHop h = OptTreeConverter.getAbstractPlanMapping().getMappedHop(n.getID());\nString inMatrix = h.getInput().get(0).getName();\n- if( cand.containsKey(inMatrix) ) //Required Condition: partitioning applicable\n+ if( cand.containsKey(inMatrix) && h.getDataType().isMatrix() ) //Required: partitionable\n{\nPartitionFormat dpf = cand.get(inMatrix);\ndouble mnew = getNewRIXMemoryEstimate( n, inMatrix, dpf, vars );\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysml/test/integration/functions/frame/ParforFrameIntermediateTest.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.sysml.test.integration.functions.frame;\n+\n+import org.apache.sysml.api.DMLScript;\n+import org.apache.sysml.api.DMLScript.RUNTIME_PLATFORM;\n+import org.apache.sysml.lops.LopProperties.ExecType;\n+import org.apache.sysml.runtime.io.FrameWriterFactory;\n+import org.apache.sysml.runtime.matrix.data.FrameBlock;\n+import org.apache.sysml.runtime.matrix.data.OutputInfo;\n+import org.apache.sysml.runtime.util.DataConverter;\n+import org.apache.sysml.test.integration.AutomatedTestBase;\n+import org.apache.sysml.test.integration.TestConfiguration;\n+import org.apache.sysml.test.utils.TestUtils;\n+import org.junit.Test;\n+\n+public class ParforFrameIntermediateTest extends AutomatedTestBase\n+{\n+ private final static String TEST_DIR = \"functions/frame/\";\n+ private final static String TEST_NAME = \"ParforFrameIntermediates\";\n+ private final static String TEST_CLASS_DIR = TEST_DIR + ParforFrameIntermediateTest.class.getSimpleName() + \"/\";\n+\n+ private final static int rows = 1382;\n+ private final static int cols = 5;\n+\n+ @Override\n+ public void setUp() {\n+ TestUtils.clearAssertionInformation();\n+ addTestConfiguration(TEST_NAME, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME, new String[] {\"F2\"}));\n+ }\n+\n+ @Test\n+ public void testParforFrameIntermediatesCP() {\n+ runParforFrameIntermediatesTest(ExecType.CP);\n+ }\n+\n+ @Test\n+ public void testParforFrameIntermediatesSpark() {\n+ runParforFrameIntermediatesTest(ExecType.SPARK);\n+ }\n+\n+ private void runParforFrameIntermediatesTest( ExecType et ) {\n+ RUNTIME_PLATFORM platformOld = rtplatform;\n+ switch( et ){\n+ case SPARK: rtplatform = RUNTIME_PLATFORM.SPARK; break;\n+ default: rtplatform = RUNTIME_PLATFORM.HYBRID_SPARK; break;\n+ }\n+\n+ boolean sparkConfigOld = DMLScript.USE_LOCAL_SPARK_CONFIG;\n+ if( rtplatform == RUNTIME_PLATFORM.SPARK\n+ || rtplatform == RUNTIME_PLATFORM.HYBRID_SPARK )\n+ DMLScript.USE_LOCAL_SPARK_CONFIG = true;\n+\n+ try\n+ {\n+ //setup testcase\n+ getAndLoadTestConfiguration(TEST_NAME);\n+ String HOME = SCRIPT_DIR + TEST_DIR;\n+ fullDMLScriptName = HOME + TEST_NAME + \".dml\";\n+ programArgs = new String[]{\"-explain\", \"-args\", input(\"F\")};\n+\n+ //generate input data and write as frame\n+ double[][] A = getRandomMatrix(rows, cols, -10, 10, 0.9, 8362);\n+ FrameBlock fA = DataConverter.convertToFrameBlock(\n+ DataConverter.convertToMatrixBlock(A));\n+ FrameWriterFactory.createFrameWriter(OutputInfo.CSVOutputInfo)\n+ .writeFrameToHDFS(fA, input(\"F\"), rows, cols);\n+\n+ //run test\n+ runTest(true, false, null, -1);\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" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/frame/ParforFrameIntermediates.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 = read($1, data_type=\"frame\", format=\"csv\");\n+\n+parfor(i in 1:ncol(F)) {\n+ Fi = F[, i];\n+ Mi = as.matrix(Fi) + 7;\n+ Fi2 = as.frame(Mi);\n+ print(toString(Fi2[1:2,]));\n+}\n" }, { "change_type": "MODIFY", "old_path": "src/test_suites/java/org/apache/sysml/test/integration/functions/frame/ZPackageSuite.java", "new_path": "src/test_suites/java/org/apache/sysml/test/integration/functions/frame/ZPackageSuite.java", "diff": "@@ -45,6 +45,7 @@ import org.junit.runners.Suite;\nFrameScalarCastingTest.class,\nFrameSchemaReadTest.class,\nFrameSerializationTest.class,\n+ ParforFrameIntermediateTest.class,\n})\n" } ]

[ { "change_type": "MODIFY", "old_path": "samples/jupyter-notebooks/DML Tips and Tricks (aka Fun With DML).ipynb", "new_path": "samples/jupyter-notebooks/DML Tips and Tricks (aka Fun With DML).ipynb", "diff": "\"cell_type\": \"markdown\",\n\"metadata\": {},\n\"source\": [\n- \"1. [Value-based join of two Matrices](#JoinMatrices)\\n\",\n+ \"1. [Cross Validation](#CrossValidation)\\n\",\n+ \"* [Value-based join of two Matrices](#JoinMatrices)\\n\",\n\"* [Filter Matrix to include only Frequent Column Values](#FilterMatrix)\\n\",\n\"* [Construct (sparse) Matrix from (rowIndex, colIndex, values) triplets](#Construct_sparse_Matrix)\\n\",\n\"* [Find and remove duplicates in columns or rows](#Find_and_remove_duplicates)\\n\",\n},\n{\n\"cell_type\": \"code\",\n- \"execution_count\": null,\n+ \"execution_count\": 2,\n\"metadata\": {\n\"collapsed\": false,\n\"scrolled\": true\n},\n- \"outputs\": [],\n+ \"outputs\": [\n+ {\n+ \"name\": \"stdout\",\n+ \"output_type\": \"stream\",\n+ \"text\": [\n+ \"2017-08-18 21:33:18 UTC\\n\"\n+ ]\n+ }\n+ ],\n\"source\": [\n\"from systemml import MLContext, dml, jvm_stdout\\n\",\n\"ml = MLContext(sc)\\n\",\n\"print (ml.buildTime())\"\n]\n},\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"## Cross Validation<a id=\\\"CrossValidation\\\" />\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"Perform kFold cross validation by running in parallel fold creation, training algorithm, test algorithm, and evaluation.\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 4,\n+ \"metadata\": {\n+ \"collapsed\": false\n+ },\n+ \"outputs\": [\n+ {\n+ \"name\": \"stdout\",\n+ \"output_type\": \"stream\",\n+ \"text\": [\n+ \"Test data Xyi2\\n\",\n+ \"10.000 11.000 12.000 4.000\\n\",\n+ \"16.000 17.000 18.000 6.000\\n\",\n+ \"\\n\",\n+ \"Train data Xyni2\\n\",\n+ \"1.000 2.000 3.000 1.000\\n\",\n+ \"4.000 5.000 6.000 2.000\\n\",\n+ \"7.000 8.000 9.000 3.000\\n\",\n+ \"13.000 14.000 15.000 5.000\\n\",\n+ \"\\n\",\n+ \"w2\\n\",\n+ \"95.000\\n\",\n+ \"106.000\\n\",\n+ \"117.000\\n\",\n+ \"\\n\",\n+ \"stats2\\n\",\n+ \"8938.000\\n\",\n+ \"\\n\",\n+ \"\\n\",\n+ \"Test data Xyi3\\n\",\n+ \"1.000 2.000 3.000 1.000\\n\",\n+ \"7.000 8.000 9.000 3.000\\n\",\n+ \"\\n\",\n+ \"Train data Xyni3\\n\",\n+ \"4.000 5.000 6.000 2.000\\n\",\n+ \"10.000 11.000 12.000 4.000\\n\",\n+ \"13.000 14.000 15.000 5.000\\n\",\n+ \"16.000 17.000 18.000 6.000\\n\",\n+ \"\\n\",\n+ \"w3\\n\",\n+ \"209.000\\n\",\n+ \"226.000\\n\",\n+ \"243.000\\n\",\n+ \"\\n\",\n+ \"stats3\\n\",\n+ \"6844.000\\n\",\n+ \"\\n\",\n+ \"\\n\",\n+ \"Test data Xyi1\\n\",\n+ \"4.000 5.000 6.000 2.000\\n\",\n+ \"13.000 14.000 15.000 5.000\\n\",\n+ \"\\n\",\n+ \"Train data Xyni1\\n\",\n+ \"1.000 2.000 3.000 1.000\\n\",\n+ \"7.000 8.000 9.000 3.000\\n\",\n+ \"10.000 11.000 12.000 4.000\\n\",\n+ \"16.000 17.000 18.000 6.000\\n\",\n+ \"\\n\",\n+ \"w1\\n\",\n+ \"158.000\\n\",\n+ \"172.000\\n\",\n+ \"186.000\\n\",\n+ \"\\n\",\n+ \"stats1\\n\",\n+ \"9853.000\\n\",\n+ \"\\n\",\n+ \"\\n\",\n+ \"SV selection vector:\\n\",\n+ \"3.000\\n\",\n+ \"1.000\\n\",\n+ \"3.000\\n\",\n+ \"2.000\\n\",\n+ \"1.000\\n\",\n+ \"2.000\\n\",\n+ \"\\n\",\n+ \"SystemML Statistics:\\n\",\n+ \"Total execution time:\\t\\t0.024 sec.\\n\",\n+ \"Number of executed Spark inst:\\t0.\\n\",\n+ \"\\n\",\n+ \"\\n\"\n+ ]\n+ }\n+ ],\n+ \"source\": [\n+ \"prog = \\\"\\\"\\\"\\n\",\n+ \"holdOut = 1/3\\n\",\n+ \"kFolds = 1/holdOut\\n\",\n+ \"\\n\",\n+ \"nRows = 6; nCols = 3; \\n\",\n+ \"\\n\",\n+ \"X = matrix(seq(1, nRows * nCols), rows = nRows, cols = nCols) # X data\\n\",\n+ \"y = matrix(seq(1, nRows), rows = nRows, cols = 1) # y label data\\n\",\n+ \"Xy = cbind (X,y) # Xy Data for CV\\n\",\n+ \"\\n\",\n+ \"sv = rand (rows = nRows, cols = 1, min = 0.0, max = 1.0, pdf = \\\"uniform\\\") # sv selection vector for fold creation \\n\",\n+ \"sv = (order(target=sv, by=1, index.return=TRUE)) %% kFolds + 1 # with numbers between 1 .. kFolds \\n\",\n+ \"\\n\",\n+ \"stats = matrix(0, rows=kFolds, cols=1) # stats per kFolds model on test data\\n\",\n+ \"\\n\",\n+ \"parfor (i in 1:kFolds)\\n\",\n+ \"{\\n\",\n+ \" # Skip empty training data or test data. \\n\",\n+ \" if ( sum (sv == i) > 0 & sum (sv == i) < nrow(X) ) \\n\",\n+ \" {\\n\",\n+ \" Xyi = removeEmpty(target = Xy, margin = \\\"rows\\\", select = (sv == i)) # Xyi fold, i.e. 1/k of rows (test data)\\n\",\n+ \" Xyni = removeEmpty(target = Xy, margin = \\\"rows\\\", select = (sv != i)) # Xyni data, i.e. (k-1)/k of rows (train data)\\n\",\n+ \"\\n\",\n+ \" # Skip extreme label inbalance\\n\",\n+ \" distinctLabels = aggregate( target = Xyni[,1], groups = Xyni[,1], fn = \\\"count\\\")\\n\",\n+ \" if ( nrow(distinctLabels) > 1)\\n\",\n+ \" {\\n\",\n+ \" wi = trainAlg (Xyni[ ,1:ncol(Xy)-1], Xyni[ ,ncol(Xy)]) # wi Model for i-th training data\\n\",\n+ \" pi = testAlg (Xyi [ ,1:ncol(Xy)-1], wi) # pi Prediction for i-th test data\\n\",\n+ \" ei = evalPrediction (pi, Xyi[ ,ncol(Xy)]) # stats[i,] evaluation of prediction of i-th fold\\n\",\n+ \" stats[i,] = ei\\n\",\n+ \" \\n\",\n+ \" print ( \\\"Test data Xyi\\\" + i + \\\"\\\\n\\\" + toString(Xyi) \\n\",\n+ \" + \\\"\\\\nTrain data Xyni\\\" + i + \\\"\\\\n\\\" + toString(Xyni) \\n\",\n+ \" + \\\"\\\\nw\\\" + i + \\\"\\\\n\\\" + toString(wi) \\n\",\n+ \" + \\\"\\\\nstats\\\" + i + \\\"\\\\n\\\" + toString(stats[i,]) \\n\",\n+ \" + \\\"\\\\n\\\")\\n\",\n+ \" }\\n\",\n+ \" else\\n\",\n+ \" {\\n\",\n+ \" print (\\\"Training data for fold \\\" + i + \\\" has only \\\" + nrow(distinctLabels) + \\\" distinct labels. Needs to be > 1.\\\")\\n\",\n+ \" } \\n\",\n+ \" } \\n\",\n+ \" else \\n\",\n+ \" {\\n\",\n+ \" print (\\\"Training data or test data for fold \\\" + i + \\\" is empty. Fold not validated.\\\")\\n\",\n+ \" }\\n\",\n+ \"\\n\",\n+ \"}\\n\",\n+ \"\\n\",\n+ \"print (\\\"SV selection vector:\\\\n\\\" + toString(sv))\\n\",\n+ \"\\n\",\n+ \"trainAlg = function (matrix[double] X, matrix[double] y)\\n\",\n+ \" return (matrix[double] w)\\n\",\n+ \"{\\n\",\n+ \" w = t(X) %*% y\\n\",\n+ \"}\\n\",\n+ \"\\n\",\n+ \"testAlg = function (matrix[double] X, matrix[double] w)\\n\",\n+ \" return (matrix[double] p)\\n\",\n+ \"{\\n\",\n+ \" p = X %*% w\\n\",\n+ \"}\\n\",\n+ \"\\n\",\n+ \"evalPrediction = function (matrix[double] p, matrix[double] y)\\n\",\n+ \" return (matrix[double] e)\\n\",\n+ \"{\\n\",\n+ \" e = as.matrix(sum (p - y))\\n\",\n+ \"}\\n\",\n+ \"\\\"\\\"\\\"\\n\",\n+ \"\\n\",\n+ \"with jvm_stdout(True):\\n\",\n+ \" ml.execute(dml(prog))\"\n+ ]\n+ },\n{\n\"cell_type\": \"markdown\",\n\"metadata\": {},\n" } ]

[ { "change_type": "MODIFY", "old_path": "dev/release/src/test/bin/verifyBuild.sh", "new_path": "dev/release/src/test/bin/verifyBuild.sh", "diff": "@@ -90,6 +90,7 @@ if [ -z $WORKING_DIR ] ; then\nWORKING_DIR=\"$USER_DIR/tmp/relValidation\"\nfi\n+rm -rf \"$WORKING_DIR\"/systemml\nmkdir -p \"$WORKING_DIR\"\nOUT_FILE=$WORKING_DIR/relValidation.out\nERR_FILE=$WORKING_DIR/relValidation.err\n@@ -171,6 +172,7 @@ runCommand \"cd ../../\"\necho \"`date +%Y-%m-%dT%H:%M:%S`: INFO: Verifying Python scripts...\"\necho \"`date +%Y-%m-%dT%H:%M:%S`: INFO: Verifying Python scripts...\" >> $OUT_FILE\nrunCommand \"pip install --upgrade systemml-$VER_NAME-python.tgz\"\n+runCommand \"pip3 install --upgrade systemml-$VER_NAME-python.tgz\"\nrunCommand \"cd ../../../\"\nrunCommand \"$SPARK_HOME/bin/spark-submit src/test/python/matrix_sum_example.py\"\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/opt/PlanSelectionFuseCostBased.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/opt/PlanSelectionFuseCostBased.java", "diff": "@@ -57,6 +57,7 @@ import org.apache.sysml.hops.codegen.template.TemplateBase.TemplateType;\nimport org.apache.sysml.hops.rewrite.HopRewriteUtils;\nimport org.apache.sysml.runtime.controlprogram.parfor.stat.InfrastructureAnalyzer;\nimport org.apache.sysml.runtime.controlprogram.parfor.util.IDSequence;\n+import org.apache.sysml.runtime.util.UtilFunctions;\nimport org.apache.sysml.utils.Statistics;\n/**\n@@ -87,12 +88,14 @@ public class PlanSelectionFuseCostBased extends PlanSelection\nCollection<PlanPartition> parts = PlanAnalyzer.analyzePlanPartitions(memo, roots, false);\n//step 2: optimize individual plan partitions\n+ int sumMatPoints = 0;\nfor( PlanPartition part : parts ) {\n//create composite templates (within the partition)\ncreateAndAddMultiAggPlans(memo, part.getPartition(), part.getRoots());\n//plan enumeration and plan selection\nselectPlans(memo, part.getPartition(), part.getRoots(), part.getMatPoints());\n+ sumMatPoints += part.getMatPoints().size();\n}\n//step 3: add composite templates (across partitions)\n@@ -101,6 +104,10 @@ public class PlanSelectionFuseCostBased extends PlanSelection\n//take all distinct best plans\nfor( Entry<Long, List<MemoTableEntry>> e : getBestPlans().entrySet() )\nmemo.setDistinct(e.getKey(), e.getValue());\n+\n+ //maintain statistics\n+ if( DMLScript.STATISTICS )\n+ Statistics.incrementCodegenEnumAll(UtilFunctions.pow(2, sumMatPoints));\n}\n//within-partition multi-agg templates\n@@ -389,7 +396,7 @@ public class PlanSelectionFuseCostBased extends PlanSelection\n}\nif( DMLScript.STATISTICS ) {\n- Statistics.incrementCodegenEnumAll(len);\n+ Statistics.incrementCodegenEnumAllP(len);\nStatistics.incrementCodegenEnumEval(len);\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/opt/PlanSelectionFuseCostBasedV2.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/opt/PlanSelectionFuseCostBasedV2.java", "diff": "@@ -107,12 +107,14 @@ public class PlanSelectionFuseCostBasedV2 extends PlanSelection\nCollection<PlanPartition> parts = PlanAnalyzer.analyzePlanPartitions(memo, roots, true);\n//step 2: optimize individual plan partitions\n+ int sumMatPoints = 0;\nfor( PlanPartition part : parts ) {\n//create composite templates (within the partition)\ncreateAndAddMultiAggPlans(memo, part.getPartition(), part.getRoots());\n//plan enumeration and plan selection\nselectPlans(memo, part);\n+ sumMatPoints += part.getMatPointsExt().length;\n}\n//step 3: add composite templates (across partitions)\n@@ -121,6 +123,10 @@ public class PlanSelectionFuseCostBasedV2 extends PlanSelection\n//take all distinct best plans\nfor( Entry<Long, List<MemoTableEntry>> e : getBestPlans().entrySet() )\nmemo.setDistinct(e.getKey(), e.getValue());\n+\n+ //maintain statistics\n+ if( DMLScript.STATISTICS )\n+ Statistics.incrementCodegenEnumAll(UtilFunctions.pow(2, sumMatPoints));\n}\nprivate void selectPlans(CPlanMemoTable memo, PlanPartition part)\n@@ -257,7 +263,7 @@ public class PlanSelectionFuseCostBasedV2 extends PlanSelection\n}\nif( DMLScript.STATISTICS ) {\n- Statistics.incrementCodegenEnumAll((rgraph!=null)?len:0);\n+ Statistics.incrementCodegenEnumAllP((rgraph!=null)?len:0);\nStatistics.incrementCodegenEnumEval(numEvalPlans);\nStatistics.incrementCodegenEnumEvalP(numEvalPartPlans);\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/utils/Statistics.java", "new_path": "src/main/java/org/apache/sysml/utils/Statistics.java", "diff": "@@ -79,6 +79,7 @@ public class Statistics\nprivate static final LongAdder codegenCPlanCompile = new LongAdder(); //count\nprivate static final LongAdder codegenClassCompile = new LongAdder(); //count\nprivate static final LongAdder codegenEnumAll = new LongAdder(); //count\n+ private static final LongAdder codegenEnumAllP = new LongAdder(); //count\nprivate static final LongAdder codegenEnumEval = new LongAdder(); //count\nprivate static final LongAdder codegenEnumEvalP = new LongAdder(); //count\nprivate static final LongAdder codegenPlanCacheHits = new LongAdder(); //count\n@@ -262,6 +263,9 @@ public class Statistics\npublic static void incrementCodegenEnumAll(long delta) {\ncodegenEnumAll.add(delta);\n}\n+ public static void incrementCodegenEnumAllP(long delta) {\n+ codegenEnumAllP.add(delta);\n+ }\npublic static void incrementCodegenEnumEval(long delta) {\ncodegenEnumEval.add(delta);\n}\n@@ -300,6 +304,9 @@ public class Statistics\npublic static long getCodegenEnumAll() {\nreturn codegenEnumAll.longValue();\n}\n+ public static long getCodegenEnumAllP() {\n+ return codegenEnumAllP.longValue();\n+ }\npublic static long getCodegenEnumEval() {\nreturn codegenEnumEval.longValue();\n}\n@@ -403,6 +410,7 @@ public class Statistics\ncodegenCPlanCompile.reset();\ncodegenClassCompile.reset();\ncodegenEnumAll.reset();\n+ codegenEnumAllP.reset();\ncodegenEnumEval.reset();\ncodegenEnumEvalP.reset();\ncodegenCompileTime.reset();\n@@ -795,8 +803,8 @@ public class Statistics\nif( ConfigurationManager.isCodegenEnabled() ) {\nsb.append(\"Codegen compile (DAG,CP,JC):\\t\" + getCodegenDAGCompile() + \"/\"\n+ getCodegenCPlanCompile() + \"/\" + getCodegenClassCompile() + \".\\n\");\n- sb.append(\"Codegen enum (All,Eval,EvalP):\\t\" + getCodegenEnumAll() + \"/\"\n- + getCodegenEnumEval() + \"/\" + getCodegenEnumEvalP() + \".\\n\");\n+ sb.append(\"Codegen enum (ALLt/p,EVALt/p):\\t\" + getCodegenEnumAll() + \"/\" +\n+ getCodegenEnumAllP() + \"/\" + getCodegenEnumEval() + \"/\" + getCodegenEnumEvalP() + \".\\n\");\nsb.append(\"Codegen compile times (DAG,JC):\\t\" + String.format(\"%.3f\", (double)getCodegenCompileTime()/1000000000) + \"/\" +\nString.format(\"%.3f\", (double)getCodegenClassCompileTime()/1000000000) + \" sec.\\n\");\nsb.append(\"Codegen plan cache hits:\\t\" + getCodegenPlanCacheHits() + \"/\" + getCodegenPlanCacheTotal() + \".\\n\");\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/ExecutionContext.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/ExecutionContext.java", "diff": "@@ -374,7 +374,7 @@ public class ExecutionContext {\n}\n// The lock is added here for an output block\n// so that any block currently in use is not deallocated by eviction on the GPU\n- mo.getGPUObject(getGPUContext(0)).addLock();\n+ mo.getGPUObject(getGPUContext(0)).addWriteLock();\nreturn mo;\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/GPUInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/GPUInstruction.java", "diff": "package org.apache.sysml.runtime.instructions.gpu;\n+import org.apache.commons.logging.Log;\n+import org.apache.commons.logging.LogFactory;\nimport org.apache.sysml.api.DMLScript;\nimport org.apache.sysml.lops.runtime.RunMRJobs;\nimport org.apache.sysml.runtime.DMLRuntimeException;\n@@ -26,6 +28,7 @@ import org.apache.sysml.runtime.controlprogram.caching.MatrixObject;\nimport org.apache.sysml.runtime.controlprogram.context.ExecutionContext;\nimport org.apache.sysml.runtime.instructions.GPUInstructionParser;\nimport org.apache.sysml.runtime.instructions.Instruction;\n+import org.apache.sysml.runtime.instructions.gpu.context.GPUContext;\nimport org.apache.sysml.runtime.matrix.data.Pair;\nimport org.apache.sysml.runtime.matrix.operators.Operator;\nimport org.apache.sysml.utils.GPUStatistics;\n@@ -47,6 +50,8 @@ public abstract class GPUInstruction extends Instruction {\nMatrixIndexing\n};\n+ private static final Log LOG = LogFactory.getLog(GPUInstruction.class.getName());\n+\n// Memory/conversions\npublic final static String MISC_TIMER_HOST_TO_DEVICE = \"H2D\"; // time spent in bringing data to gpu (from host)\npublic final static String MISC_TIMER_DEVICE_TO_HOST = \"D2H\"; // time spent in bringing data from gpu (to host)\n@@ -191,6 +196,13 @@ public abstract class GPUInstruction extends Instruction {\nif(DMLScript.SYNCHRONIZE_GPU) {\njcuda.runtime.JCuda.cudaDeviceSynchronize();\n}\n+ if(LOG.isDebugEnabled()) {\n+ for(GPUContext gpuCtx : ec.getGPUContexts()) {\n+ if(gpuCtx != null)\n+ gpuCtx.printMemoryInfo(getOpcode());\n+ }\n+ }\n+\n}\n/**\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/context/GPUContext.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/context/GPUContext.java", "diff": "@@ -43,6 +43,7 @@ import java.util.Comparator;\nimport java.util.HashMap;\nimport java.util.LinkedList;\nimport java.util.Map;\n+import java.util.Map.Entry;\nimport org.apache.commons.logging.Log;\nimport org.apache.commons.logging.LogFactory;\n@@ -131,6 +132,40 @@ public class GPUContext {\n*/\nprivate final ThreadLocal<JCudaKernels> kernels = new ThreadLocal<>();\n+ /**\n+ * Print information of memory usage.\n+ *\n+ * @param opcode opcode of caller\n+ * @throws DMLRuntimeException if error\n+ */\n+ public void printMemoryInfo(String opcode) throws DMLRuntimeException {\n+ if(LOG.isDebugEnabled()) {\n+ long totalFreeCUDASpace = 0;\n+ for(Entry<Long, LinkedList<Pointer>> kv : freeCUDASpaceMap.entrySet()) {\n+ totalFreeCUDASpace += kv.getKey()*kv.getValue().size();\n+ }\n+ long readLockedAllocatedMemory = 0;\n+ long writeLockedAllocatedMemory = 0;\n+ long unlockedAllocatedMemory = 0;\n+ for(GPUObject gpuObj : allocatedGPUObjects) {\n+ if(gpuObj.readLocks.longValue() > 0)\n+ readLockedAllocatedMemory += gpuObj.getSizeOnDevice();\n+ else if(gpuObj.writeLock)\n+ writeLockedAllocatedMemory += gpuObj.getSizeOnDevice();\n+ else\n+ unlockedAllocatedMemory += gpuObj.getSizeOnDevice();\n+ }\n+ long free[] = { 0 };\n+ long total[] = { 0 };\n+ cudaMemGetInfo(free, total);\n+ long gpuFreeMemory = (long) (free[0] * GPU_MEMORY_UTILIZATION_FACTOR);\n+ LOG.debug(opcode + \": Total memory: \" + total[0] + \", Free memory: \" + free[0] + \" (with util factor: \" + gpuFreeMemory + \"), \"\n+ + \"Lazy unfreed memory: \" + totalFreeCUDASpace + \", Locked allocated memory (read/write): \"\n+ + readLockedAllocatedMemory + \"/\" + writeLockedAllocatedMemory + \", \"\n+ + \" Unlocked allocated memory: \" + unlockedAllocatedMemory);\n+ }\n+ }\n+\nprotected GPUContext(int deviceNum) throws DMLRuntimeException {\nthis.deviceNum = deviceNum;\ncudaSetDevice(deviceNum);\n@@ -472,18 +507,19 @@ public class GPUContext {\nCollections.sort(allocatedGPUObjects, new Comparator<GPUObject>() {\n@Override\npublic int compare(GPUObject p1, GPUObject p2) {\n- long p1Val = p1.locks.get();\n- long p2Val = p2.locks.get();\n-\n- if (p1Val > 0 && p2Val > 0) {\n+ if (p1.isLocked() && p2.isLocked()) {\n// Both are locked, so don't sort\nreturn 0;\n- } else if (p1Val > 0 || p2Val > 0) {\n+ } else if (p1.isLocked()) {\n+ // Put the unlocked one to RHS\n+ // a value less than 0 if x < y; and a value greater than 0 if x > y\n+ return -1;\n+ } else if (p2.isLocked()) {\n// Put the unlocked one to RHS\n- return Long.compare(p2Val, p1Val);\n+ // a value less than 0 if x < y; and a value greater than 0 if x > y\n+ return 1;\n} else {\n// Both are unlocked\n-\nif (evictionPolicy == EvictionPolicy.MIN_EVICT) {\nlong p1Size = 0;\nlong p2Size = 0;\n@@ -510,7 +546,7 @@ public class GPUContext {\nwhile (neededSize > getAvailableMemory() && allocatedGPUObjects.size() > 0) {\nGPUObject toBeRemoved = allocatedGPUObjects.get(allocatedGPUObjects.size() - 1);\n- if (toBeRemoved.locks.get() > 0) {\n+ if (toBeRemoved.isLocked()) {\nthrow new DMLRuntimeException(\n\"There is not enough memory on device for this matrix, request (\" + neededSize + \"). Allocated GPU objects:\" + allocatedGPUObjects.toString());\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/context/GPUObject.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/context/GPUObject.java", "diff": "@@ -34,6 +34,7 @@ import static jcuda.runtime.cudaMemcpyKind.cudaMemcpyHostToDevice;\nimport java.util.Arrays;\nimport java.util.concurrent.atomic.AtomicLong;\n+import java.util.concurrent.atomic.LongAdder;\nimport org.apache.commons.logging.Log;\nimport org.apache.commons.logging.LogFactory;\n@@ -99,9 +100,14 @@ public class GPUObject {\nprotected boolean dirty = false;\n/**\n- * number of read/write locks on this object (this GPUObject is being used in a current instruction)\n+ * number of read locks on this object (this GPUObject is being used in a current instruction)\n*/\n- protected AtomicLong locks = new AtomicLong();\n+ protected LongAdder readLocks = new LongAdder();\n+\n+ /**\n+ * whether write lock on this object (this GPUObject is being used in a current instruction)\n+ */\n+ protected boolean writeLock = false;\n/**\n* Timestamp, needed by {@link GPUContext#evict(long)}\n@@ -132,7 +138,11 @@ public class GPUObject {\nthat.allocateTensorDescriptor(me.tensorShape[0], me.tensorShape[1], me.tensorShape[2], me.tensorShape[3]);\n}\nthat.dirty = me.dirty;\n- that.locks = new AtomicLong(me.locks.get());\n+ // TODO Nakul: Should the locks be cloned here ?\n+ // The only place clone is getting called: LibMatrixCUDA's solve\n+ that.readLocks.reset();\n+ that.writeLock = false;\n+\nthat.timestamp = new AtomicLong(me.timestamp.get());\nthat.isSparse = me.isSparse;\n@@ -618,7 +628,7 @@ public class GPUObject {\ncopyFromHostToDevice(opcode);\ntransferred = true;\n}\n- addLock();\n+ addReadLock();\nif (!isAllocated())\nthrow new DMLRuntimeException(\"Expected device data to be allocated\");\nreturn transferred;\n@@ -664,10 +674,6 @@ public class GPUObject {\nreturn allocated;\n}\n- public void addLock() {\n- locks.addAndGet(1);\n- }\n-\n/**\n* if the data is allocated on the GPU and is dirty, it is copied back to the host memory\n*\n@@ -694,21 +700,50 @@ public class GPUObject {\nreturn copied;\n}\n+ public boolean isLocked() {\n+ return writeLock || readLocks.longValue() > 0;\n+ }\n+\n+ public void addReadLock() throws DMLRuntimeException {\n+ if(writeLock)\n+ throw new DMLRuntimeException(\"Attempting to add a read lock when writeLock=\"+ writeLock);\n+ else\n+ readLocks.increment();\n+ }\n+\n+ public void addWriteLock() throws DMLRuntimeException {\n+ if(readLocks.longValue() > 0)\n+ throw new DMLRuntimeException(\"Attempting to add a write lock when readLocks=\"+ readLocks.longValue());\n+ else if(writeLock)\n+ throw new DMLRuntimeException(\"Attempting to add a write lock when writeLock=\"+ writeLock);\n+ else\n+ writeLock = true;\n+ }\n+\n+ public void releaseReadLock() throws DMLRuntimeException {\n+ readLocks.decrement();\n+ if(readLocks.longValue() < 0)\n+ throw new DMLRuntimeException(\"Attempting to release a read lock when readLocks=\"+ readLocks.longValue());\n+ }\n+\n+ public void releaseWriteLock() throws DMLRuntimeException {\n+ if(writeLock)\n+ writeLock = false;\n+ else\n+ throw new DMLRuntimeException(\"Internal state error : Attempting to release write lock on a GPUObject, which was already released\");\n+ }\n+\n+ public void resetReadWriteLock() {\n+ readLocks.reset();\n+ writeLock = false;\n+ }\n+\n/**\n* Updates the locks depending on the eviction policy selected\n*\n* @throws DMLRuntimeException if there is no locked GPU Object or if could not obtain a {@link GPUContext}\n*/\n- private void updateReleaseLocks(int l) throws DMLRuntimeException {\n- int newLocks = (int) locks.addAndGet(l);\n- if (newLocks < 0) {\n- throw new CacheException(\"Internal state error : Invalid number of locks on a GPUObject\");\n- }\n-\n- if(LOG.isTraceEnabled()) {\n- LOG.trace(\"GPU : updateReleaseLocks, new number of locks is \" + newLocks + \", on \" + this + \", GPUContext=\"\n- + getGPUContext());\n- }\n+ private void updateReleaseLocks() throws DMLRuntimeException {\nGPUContext.EvictionPolicy evictionPolicy = getGPUContext().evictionPolicy;\nswitch (evictionPolicy) {\ncase LRU:\n@@ -730,8 +765,8 @@ public class GPUObject {\n* @throws DMLRuntimeException if data is not allocated or if there is no locked GPU Object or if could not obtain a {@link GPUContext}\n*/\npublic void releaseInput() throws DMLRuntimeException {\n- // A read lock is a positive quantity, therefor when the lock is freed, a negative 1 is added\n- updateReleaseLocks(-1);\n+ releaseReadLock();\n+ updateReleaseLocks();\nif (!isAllocated())\nthrow new CacheException(\"Attempting to release an input before allocating it\");\n}\n@@ -742,8 +777,8 @@ public class GPUObject {\n* @throws DMLRuntimeException if data is not allocated or if there is no locked GPU Object or if could not obtain a {@link GPUContext}\n*/\npublic void releaseOutput() throws DMLRuntimeException {\n- // A write lock is a negative quantity, therefore when the lock is freed, a positive number is added\n- updateReleaseLocks(1);\n+ releaseWriteLock();\n+ updateReleaseLocks();\ndirty = true;\nif (!isAllocated())\nthrow new CacheException(\"Attempting to release an output before allocating it\");\n@@ -798,7 +833,7 @@ public class GPUObject {\ncudnnDestroyTensorDescriptor(tensorDescriptor);\ntensorDescriptor = null;\n}\n- locks.set(0);\n+ resetReadWriteLock();\ngetGPUContext().removeRecordedUsage(this);\n}\n@@ -1061,7 +1096,8 @@ public class GPUObject {\nfinal StringBuilder sb = new StringBuilder(\"GPUObject{\");\nsb.append(\", tensorShape=\").append(Arrays.toString(tensorShape));\nsb.append(\", dirty=\").append(dirty);\n- sb.append(\", locks=\").append(locks);\n+ sb.append(\", readLocks=\").append(readLocks.longValue());\n+ sb.append(\", writeLock=\").append(writeLock);\nsb.append(\", sparse? \").append(isSparse);\nsb.append(\", dims=[\").append(mat.getNumRows()).append(\",\").append(mat.getNumColumns()).append(\"]\");\nsb.append('}');\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "<version>18</version>\n</parent>\n<groupId>org.apache.systemml</groupId>\n- <version>1.0.0-SNAPSHOT</version>\n+ <version>0.1.0</version>\n<artifactId>systemml</artifactId>\n<packaging>jar</packaging>\n<name>SystemML</name>\n<connection>scm:git:[email protected]:apache/systemml</connection>\n<developerConnection>scm:git:https://git-wip-us.apache.org/repos/asf/systemml</developerConnection>\n<url>https://git-wip-us.apache.org/repos/asf?p=systemml.git</url>\n- <tag>HEAD</tag>\n+ <tag>v0.1.0-rc1</tag>\n</scm>\n<issueManagement>\n<system>JIRA</system>\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/jmlc/PreparedScript.java", "new_path": "src/main/java/org/apache/sysml/api/jmlc/PreparedScript.java", "diff": "@@ -448,14 +448,14 @@ public class PreparedScript\n//enable requested functions for recompile once\nfor( String fname : fnames ) {\nString fkey = DMLProgram.constructFunctionKey(fnamespace, fname);\n- if( !fgraph.isRecursiveFunction(fkey) ) {\n+ if( fgraph != null && !fgraph.isRecursiveFunction(fkey) ) {\nFunctionProgramBlock fpb = _prog.getFunctionProgramBlock(fnamespace, fname);\nif( fpb != null )\nfpb.setRecompileOnce(true);\nelse\nLOG.warn(\"Failed to enable function recompile for non-existing '\"+fkey+\"'.\");\n}\n- else {\n+ else if( fgraph != null ) {\nLOG.warn(\"Failed to enable function recompile for recursive '\"+fkey+\"'.\");\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/api/mlcontext/MLContextConversionUtil.java", "new_path": "src/main/java/org/apache/sysml/api/mlcontext/MLContextConversionUtil.java", "diff": "@@ -308,12 +308,14 @@ public class MLContextConversionUtil {\npublic static FrameObject binaryBlocksToFrameObject(String variableName, JavaPairRDD<Long, FrameBlock> binaryBlocks,\nFrameMetadata frameMetadata) {\n- MatrixCharacteristics mc = (frameMetadata != null) ? frameMetadata.asMatrixCharacteristics()\n- : new MatrixCharacteristics();\n+ MatrixCharacteristics mc = (frameMetadata != null) ?\n+ frameMetadata.asMatrixCharacteristics() : new MatrixCharacteristics();\n+ ValueType[] schema = (frameMetadata != null) ?\n+ frameMetadata.getFrameSchema().getSchema().toArray(new ValueType[0]) :\n+ UtilFunctions.nCopies((int)mc.getCols(), ValueType.STRING);\nFrameObject frameObject = new FrameObject(OptimizerUtils.getUniqueTempFileName(),\n- new MatrixFormatMetaData(mc, OutputInfo.BinaryBlockOutputInfo, InputInfo.BinaryBlockInputInfo),\n- frameMetadata.getFrameSchema().getSchema().toArray(new ValueType[0]));\n+ new MatrixFormatMetaData(mc, OutputInfo.BinaryBlockOutputInfo, InputInfo.BinaryBlockInputInfo), schema);\nframeObject.setRDDHandle(new RDDObject(binaryBlocks, variableName));\nreturn frameObject;\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/UnaryOp.java", "new_path": "src/main/java/org/apache/sysml/hops/UnaryOp.java", "diff": "@@ -584,7 +584,8 @@ public class UnaryOp extends Hop implements MultiThreadedHop\n}\nif (isGPUEnabled()) {\n- OptimizerUtils.estimateSize(dim1, dim2); // Intermediate memory required to convert sparse to dense\n+ // Intermediate memory required to convert sparse to dense\n+ ret += OptimizerUtils.estimateSize(dim1, dim2);\n}\nreturn ret;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateCell.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateCell.java", "diff": "@@ -366,7 +366,7 @@ public class TemplateCell extends TemplateBase\nif( h1.isScalar() && h2.isScalar() )\nreturn Long.compare(h1.getHopID(), h2.getHopID());\nreturn (h1.dimsKnown(true) && h2.dimsKnown(true) && h1.getNnz() != h2.getNnz()\n- && HopRewriteUtils.isSparse(h1) || HopRewriteUtils.isSparse(h1)) ?\n+ && (HopRewriteUtils.isSparse(h1) || HopRewriteUtils.isSparse(h2))) ?\nLong.compare(h1.getNnz(), h2.getNnz()) :\nLong.compare(h1.getHopID(), h2.getHopID());\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/MatrixAppendGPUInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/MatrixAppendGPUInstruction.java", "diff": "@@ -39,7 +39,7 @@ public class MatrixAppendGPUInstruction extends GPUInstruction {\nCPOperand output;\nCPOperand input1, input2;\n- AppendCPInstruction.AppendType type;\n+ AppendCPInstruction.AppendType atype;\nprivate MatrixAppendGPUInstruction(Operator op, CPOperand in1, CPOperand in2, CPOperand out,\nAppendCPInstruction.AppendType type, String opcode, String istr) {\n@@ -47,7 +47,7 @@ public class MatrixAppendGPUInstruction extends GPUInstruction {\nthis.output = out;\nthis.input1 = in1;\nthis.input2 = in2;\n- this.type = type;\n+ this.atype = type;\n}\npublic static MatrixAppendGPUInstruction parseInstruction ( String str )\n@@ -86,9 +86,9 @@ public class MatrixAppendGPUInstruction extends GPUInstruction {\nMatrixObject mat1 = getMatrixInputForGPUInstruction(ec, input1.getName());\nMatrixObject mat2 = getMatrixInputForGPUInstruction(ec, input2.getName());\n- if(type == AppendCPInstruction.AppendType.CBIND) {\n+ if(atype == AppendCPInstruction.AppendType.CBIND) {\nLibMatrixCUDA.cbind(ec, ec.getGPUContext(0), getExtendedOpcode(), mat1, mat2, output.getName());\n- } else if (type == AppendCPInstruction.AppendType.RBIND ) {\n+ } else if (atype == AppendCPInstruction.AppendType.RBIND ) {\nLibMatrixCUDA.rbind(ec, ec.getGPUContext(0), getExtendedOpcode(), mat1, mat2, output.getName());\n} else {\nthrow new DMLRuntimeException(\"Unsupported GPU operator:\" + opcode);\n" } ]

[ { "change_type": "MODIFY", "old_path": "dev/release/release-build.sh", "new_path": "dev/release/release-build.sh", "diff": "@@ -265,11 +265,14 @@ if [[ \"$RELEASE_PREPARE\" == \"true\" ]]; then\necho \"RELEASE_STAGING_LOCATION=$RELEASE_STAGING_LOCATION\"\necho \"BASE_DIR=$BASE_DIR\"\n- exit 5\n+ # As fix has been added below to update version information exit to update pom file is not needed.\n+ # exit 5\n# Update dev/release/target/release/systemml/pom.xml with similar to following contents which is for 0.13.0 RC1\n# Update <version>0.13.0</version>\n# Update <tag>v0.13.0-rc1</tag>\n+ sed -i .bak \"s|<version>$DEVELOPMENT_VERSION<\\/version>|<version>$RELEASE_VERSION<\\/version>|\" $BASE_DIR/target/release/systemml/pom.xml\n+ sed -i .bak \"s|<tag>HEAD<\\/tag>|<tag>$RELEASE_TAG<\\/tag>|\" $BASE_DIR/target/release/systemml/pom.xml\ncd $RELEASE_WORK_DIR/systemml\n## Rerunning mvn with clean and package goals, as release:prepare changes ordeer for some dependencies like unpack and shade.\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/misc/RewriteElementwiseMultChainOptimizationAllTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/misc/RewriteElementwiseMultChainOptimizationAllTest.java", "diff": "@@ -82,7 +82,7 @@ public class RewriteElementwiseMultChainOptimizationAllTest extends AutomatedTes\n}\nboolean sparkConfigOld = DMLScript.USE_LOCAL_SPARK_CONFIG;\n- if( rtplatform == RUNTIME_PLATFORM.SPARK )\n+ if( rtplatform == RUNTIME_PLATFORM.SPARK || rtplatform == RUNTIME_PLATFORM.HYBRID_SPARK )\nDMLScript.USE_LOCAL_SPARK_CONFIG = true;\nboolean rewritesOld = OptimizerUtils.ALLOW_SUM_PRODUCT_REWRITES;\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/misc/RewriteElementwiseMultChainOptimizationTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/misc/RewriteElementwiseMultChainOptimizationTest.java", "diff": "@@ -81,7 +81,7 @@ public class RewriteElementwiseMultChainOptimizationTest extends AutomatedTestBa\n}\nboolean sparkConfigOld = DMLScript.USE_LOCAL_SPARK_CONFIG;\n- if( rtplatform == RUNTIME_PLATFORM.SPARK )\n+ if( rtplatform == RUNTIME_PLATFORM.SPARK || rtplatform == RUNTIME_PLATFORM.HYBRID_SPARK )\nDMLScript.USE_LOCAL_SPARK_CONFIG = true;\nboolean rewritesOld = OptimizerUtils.ALLOW_SUM_PRODUCT_REWRITES;\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/misc/RewriteMatrixMultChainOptTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/misc/RewriteMatrixMultChainOptTest.java", "diff": "@@ -78,7 +78,7 @@ public class RewriteMatrixMultChainOptTest extends AutomatedTestBase\n}\nboolean sparkConfigOld = DMLScript.USE_LOCAL_SPARK_CONFIG;\n- if( rtplatform == RUNTIME_PLATFORM.SPARK )\n+ if( rtplatform == RUNTIME_PLATFORM.SPARK || rtplatform == RUNTIME_PLATFORM.HYBRID_SPARK )\nDMLScript.USE_LOCAL_SPARK_CONFIG = true;\nboolean rewritesOld = OptimizerUtils.ALLOW_ALGEBRAIC_SIMPLIFICATION;\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/ReblockSPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/spark/ReblockSPInstruction.java", "diff": "@@ -174,8 +174,13 @@ public class ReblockSPInstruction extends UnarySPInstruction {\n//BINARY BLOCK <- BINARY BLOCK (different sizes)\nJavaPairRDD<MatrixIndexes, MatrixBlock> in1 = sec.getBinaryBlockRDDHandleForVariable(input1.getName());\n- JavaPairRDD<MatrixIndexes, MatrixBlock> out =\n- in1.flatMapToPair(new ExtractBlockForBinaryReblock(mc, mcOut));\n+ boolean shuffleFreeReblock = mc.dimsKnown() && mcOut.dimsKnown()\n+ && (mc.getRows() < mcOut.getRowsPerBlock() || mc.getRowsPerBlock()%mcOut.getRowsPerBlock() == 0)\n+ && (mc.getCols() < mcOut.getColsPerBlock() || mc.getColsPerBlock()%mcOut.getColsPerBlock() == 0);\n+\n+ JavaPairRDD<MatrixIndexes, MatrixBlock> out = in1\n+ .flatMapToPair(new ExtractBlockForBinaryReblock(mc, mcOut));\n+ if( !shuffleFreeReblock )\nout = RDDAggregateUtils.mergeByKey(out, false);\n//put output RDD handle into symbol table\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/cpp/CMakeLists.txt", "new_path": "src/main/cpp/CMakeLists.txt", "diff": "@@ -29,6 +29,7 @@ option(USE_INTEL_MKL \"Whether to use Intel MKL (Defaults to compiling with Intel\n# Build a shared libraray\nadd_library(systemml SHARED libmatrixdnn.cpp libmatrixmult.cpp systemml.cpp)\n+set_target_properties(systemml PROPERTIES MACOSX_RPATH 1)\nset(MATH_LIBRARIES \"\")\n@@ -72,5 +73,6 @@ if (USE_OPEN_BLAS)\nfind_package(OpenMP REQUIRED)\nset_target_properties(systemml PROPERTIES LINK_FLAGS \"${OpenMP_CXX_FLAGS} ${MATH_LIBRARIES}\")\nelseif(USE_INTEL_MKL)\n- set_target_properties(systemml PROPERTIES LINK_FLAGS ${MATH_LIBRARIES}\")\n+ set_target_properties(systemml PROPERTIES LINK_FLAGS \"${MATH_LIBRARIES}\")\nendif()\n+\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/context/GPUContext.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/context/GPUContext.java", "diff": "@@ -41,9 +41,11 @@ import java.util.ArrayList;\nimport java.util.Collections;\nimport java.util.Comparator;\nimport java.util.HashMap;\n-import java.util.LinkedList;\n+import java.util.HashSet;\n+import java.util.Iterator;\nimport java.util.Map;\nimport java.util.Map.Entry;\n+import java.util.Set;\nimport org.apache.commons.logging.Log;\nimport org.apache.commons.logging.LogFactory;\n@@ -88,83 +90,50 @@ public class GPUContext {\n* active device assigned to this GPUContext instance\n*/\nprivate final int deviceNum;\n- // Invoke cudaMemGetInfo to get available memory information. Useful if GPU is shared among multiple application.\n- public double GPU_MEMORY_UTILIZATION_FACTOR = ConfigurationManager.getDMLConfig()\n- .getDoubleValue(DMLConfig.GPU_MEMORY_UTILIZATION_FACTOR);\n- /**\n- * Map of free blocks allocate on GPU. maps size_of_block -> pointer on GPU\n- */\n- private LRUCacheMap<Long, LinkedList<Pointer>> freeCUDASpaceMap = new LRUCacheMap<>();\n- /**\n- * To record size of allocated blocks\n- */\n- private HashMap<Pointer, Long> cudaBlockSizeMap = new HashMap<>();\n- /**\n- * list of allocated {@link GPUObject} instances allocated on {@link GPUContext#deviceNum} GPU\n- * These are matrices allocated on the GPU on which rmvar hasn't been called yet.\n- * If a {@link GPUObject} has more than one lock on it, it cannot be freed\n- * If it has zero locks on it, it can be freed, but it is preferrable to keep it around\n- * so that an extraneous host to dev transfer can be avoided\n- */\n- private ArrayList<GPUObject> allocatedGPUObjects = new ArrayList<>();\n/**\n* cudnnHandle for Deep Neural Network operations on the GPU\n*/\n- private final ThreadLocal<cudnnHandle> cudnnHandle = new ThreadLocal<>();\n+ private cudnnHandle cudnnHandle;\n/**\n* cublasHandle for BLAS operations on the GPU\n*/\n- private final ThreadLocal<cublasHandle> cublasHandle = new ThreadLocal<>();\n+ private cublasHandle cublasHandle;\n/**\n* cusparseHandle for certain sparse BLAS operations on the GPU\n*/\n- private final ThreadLocal<cusparseHandle> cusparseHandle = new ThreadLocal<>();\n+ private cusparseHandle cusparseHandle;\n/**\n* cusolverDnHandle for invoking solve() function on dense matrices on the GPU\n*/\n- private final ThreadLocal<cusolverDnHandle> cusolverDnHandle = new ThreadLocal<>();\n+ private cusolverDnHandle cusolverDnHandle;\n/**\n* cusolverSpHandle for invoking solve() function on sparse matrices on the GPU\n*/\n- private final ThreadLocal<cusolverSpHandle> cusolverSpHandle = new ThreadLocal<>();\n+ private cusolverSpHandle cusolverSpHandle;\n/**\n* to launch custom CUDA kernel, specific to the active GPU for this GPUContext\n*/\n- private final ThreadLocal<JCudaKernels> kernels = new ThreadLocal<>();\n+ private JCudaKernels kernels;\n+ // Invoke cudaMemGetInfo to get available memory information. Useful if GPU is shared among multiple application.\n+ public double GPU_MEMORY_UTILIZATION_FACTOR = ConfigurationManager.getDMLConfig()\n+ .getDoubleValue(DMLConfig.GPU_MEMORY_UTILIZATION_FACTOR);\n/**\n- * Print information of memory usage.\n- *\n- * @param opcode opcode of caller\n- * @throws DMLRuntimeException if error\n+ * Map of free blocks allocate on GPU. maps size_of_block -> pointer on GPU\n*/\n- public void printMemoryInfo(String opcode) throws DMLRuntimeException {\n- if(LOG.isDebugEnabled()) {\n- long totalFreeCUDASpace = 0;\n- for(Entry<Long, LinkedList<Pointer>> kv : freeCUDASpaceMap.entrySet()) {\n- totalFreeCUDASpace += kv.getKey()*kv.getValue().size();\n- }\n- long readLockedAllocatedMemory = 0;\n- long writeLockedAllocatedMemory = 0;\n- long unlockedAllocatedMemory = 0;\n- for(GPUObject gpuObj : allocatedGPUObjects) {\n- if(gpuObj.readLocks.longValue() > 0)\n- readLockedAllocatedMemory += gpuObj.getSizeOnDevice();\n- else if(gpuObj.writeLock)\n- writeLockedAllocatedMemory += gpuObj.getSizeOnDevice();\n- else\n- unlockedAllocatedMemory += gpuObj.getSizeOnDevice();\n- }\n- long free[] = { 0 };\n- long total[] = { 0 };\n- cudaMemGetInfo(free, total);\n- long gpuFreeMemory = (long) (free[0] * GPU_MEMORY_UTILIZATION_FACTOR);\n- LOG.debug(opcode + \": Total memory: \" + total[0] + \", Free memory: \" + free[0] + \" (with util factor: \" + gpuFreeMemory + \"), \"\n- + \"Lazy unfreed memory: \" + totalFreeCUDASpace + \", Locked allocated memory (read/write): \"\n- + readLockedAllocatedMemory + \"/\" + writeLockedAllocatedMemory + \", \"\n- + \" Unlocked allocated memory: \" + unlockedAllocatedMemory);\n- }\n- }\n+ private LRUCacheMap<Long, Set<Pointer>> freeCUDASpaceMap = new LRUCacheMap<>();\n+ /**\n+ * To record size of allocated blocks\n+ */\n+ private HashMap<Pointer, Long> cudaBlockSizeMap = new HashMap<>();\n+ /**\n+ * list of allocated {@link GPUObject} instances allocated on {@link GPUContext#deviceNum} GPU\n+ * These are matrices allocated on the GPU on which rmvar hasn't been called yet.\n+ * If a {@link GPUObject} has more than one lock on it, it cannot be freed\n+ * If it has zero locks on it, it can be freed, but it is preferrable to keep it around\n+ * so that an extraneous host to dev transfer can be avoided\n+ */\n+ private ArrayList<GPUObject> allocatedGPUObjects = new ArrayList<>();\nprotected GPUContext(int deviceNum) throws DMLRuntimeException {\nthis.deviceNum = deviceNum;\n@@ -194,49 +163,85 @@ public class GPUContext {\n}\n}\n+ /**\n+ * Returns which device is currently being used.\n+ *\n+ * @return the current device for the calling host thread\n+ */\n+ public static int cudaGetDevice() {\n+ int[] device = new int[1];\n+ JCuda.cudaGetDevice(device);\n+ return device[0];\n+ }\n+\n+ /**\n+ * Print information of memory usage.\n+ *\n+ * @param opcode opcode of caller\n+ * @throws DMLRuntimeException if error\n+ */\n+ public void printMemoryInfo(String opcode) throws DMLRuntimeException {\n+ if (LOG.isDebugEnabled()) {\n+ long totalFreeCUDASpace = 0;\n+ for (Entry<Long, Set<Pointer>> kv : freeCUDASpaceMap.entrySet()) {\n+ totalFreeCUDASpace += kv.getKey() * kv.getValue().size();\n+ }\n+ long readLockedAllocatedMemory = 0;\n+ long writeLockedAllocatedMemory = 0;\n+ long unlockedAllocatedMemory = 0;\n+ for (GPUObject gpuObj : allocatedGPUObjects) {\n+ if (gpuObj.readLocks.longValue() > 0)\n+ readLockedAllocatedMemory += gpuObj.getSizeOnDevice();\n+ else if (gpuObj.writeLock)\n+ writeLockedAllocatedMemory += gpuObj.getSizeOnDevice();\n+ else\n+ unlockedAllocatedMemory += gpuObj.getSizeOnDevice();\n+ }\n+ long free[] = { 0 };\n+ long total[] = { 0 };\n+ cudaMemGetInfo(free, total);\n+ long gpuFreeMemory = (long) (free[0] * GPU_MEMORY_UTILIZATION_FACTOR);\n+ LOG.debug(opcode + \": Total memory: \" + total[0] + \", Free memory: \" + free[0] + \" (with util factor: \"\n+ + gpuFreeMemory + \"), \" + \"Lazy unfreed memory: \" + totalFreeCUDASpace\n+ + \", Locked allocated memory (read/write): \" + readLockedAllocatedMemory + \"/\"\n+ + writeLockedAllocatedMemory + \", \" + \" Unlocked allocated memory: \" + unlockedAllocatedMemory);\n+ }\n+ }\n+\nprivate void initializeCudaLibraryHandles() throws DMLRuntimeException {\n- if (cudnnHandle.get() == null) {\n- cudnnHandle.set(new cudnnHandle());\n- cudnnCreate(cudnnHandle.get());\n+ deleteCudaLibraryHandles();\n+\n+ if (cudnnHandle == null) {\n+ cudnnHandle = new cudnnHandle();\n+ cudnnCreate(cudnnHandle);\n}\n- if (cublasHandle.get() == null) {\n- cublasHandle.set(new cublasHandle());\n- cublasCreate(cublasHandle.get());\n+ if (cublasHandle == null) {\n+ cublasHandle = new cublasHandle();\n+ cublasCreate(cublasHandle);\n}\n// For cublas v2, cublasSetPointerMode tells Cublas whether to expect scalar arguments on device or on host\n// This applies to arguments like \"alpha\" in Dgemm, and \"y\" in Ddot.\n// cublasSetPointerMode(LibMatrixCUDA.cublasHandle, cublasPointerMode.CUBLAS_POINTER_MODE_DEVICE);\n- if (cusparseHandle.get() == null) {\n- cusparseHandle.set(new cusparseHandle());\n- cusparseCreate(cusparseHandle.get());\n+ if (cusparseHandle == null) {\n+ cusparseHandle = new cusparseHandle();\n+ cusparseCreate(cusparseHandle);\n}\n- if (cusolverDnHandle.get() == null) {\n- cusolverDnHandle.set(new cusolverDnHandle());\n- cusolverDnCreate(cusolverDnHandle.get());\n+ if (cusolverDnHandle == null) {\n+ cusolverDnHandle = new cusolverDnHandle();\n+ cusolverDnCreate(cusolverDnHandle);\n}\n- if (cusolverSpHandle.get() == null) {\n- cusolverSpHandle.set(new cusolverSpHandle());\n- cusolverSpCreate(cusolverSpHandle.get());\n+ if (cusolverSpHandle == null) {\n+ cusolverSpHandle = new cusolverSpHandle();\n+ cusolverSpCreate(cusolverSpHandle);\n}\n- if (kernels.get() == null) {\n- kernels.set(new JCudaKernels());\n- }\n+ if (kernels == null) {\n+ kernels = new JCudaKernels();\n}\n-\n- /**\n- * Returns which device is currently being used.\n- *\n- * @return the current device for the calling host thread\n- */\n- public static int cudaGetDevice() {\n- int[] device = new int[1];\n- JCuda.cudaGetDevice(device);\n- return device[0];\n}\n/**\n@@ -301,13 +306,18 @@ public class GPUContext {\nPointer A;\nif (freeCUDASpaceMap.containsKey(size)) {\nif (LOG.isTraceEnabled()) {\n- LOG.trace(\"GPU : in allocate from instruction \" + instructionName + \", found free block of size \" + (size\n+ LOG.trace(\n+ \"GPU : in allocate from instruction \" + instructionName + \", found free block of size \" + (size\n/ 1024.0) + \" Kbytes from previously allocated block on \" + this);\n}\nif (instructionName != null && GPUStatistics.DISPLAY_STATISTICS)\nt0 = System.nanoTime();\n- LinkedList<Pointer> freeList = freeCUDASpaceMap.get(size);\n- A = freeList.pop();\n+ Set<Pointer> freeList = freeCUDASpaceMap.get(size);\n+\n+ Iterator<Pointer> it = freeList.iterator(); // at this point, freeList should have at least one element\n+ A = it.next();\n+ it.remove();\n+\nif (freeList.isEmpty())\nfreeCUDASpaceMap.remove(size);\nif (instructionName != null && GPUStatistics.DISPLAY_STATISTICS)\n@@ -316,8 +326,8 @@ public class GPUContext {\n} else {\nif (LOG.isTraceEnabled()) {\nLOG.trace(\n- \"GPU : in allocate from instruction \" + instructionName + \", allocating new block of size \" + (size\n- / 1024.0) + \" Kbytes on \" + this);\n+ \"GPU : in allocate from instruction \" + instructionName + \", allocating new block of size \" + (\n+ size / 1024.0) + \" Kbytes on \" + this);\n}\nif (DMLScript.STATISTICS)\nt0 = System.nanoTime();\n@@ -336,8 +346,8 @@ public class GPUContext {\nif (DMLScript.STATISTICS)\nt1 = System.nanoTime();\nif (LOG.isTraceEnabled()) {\n- LOG.trace(\"GPU : in allocate from instruction \" + instructionName + \", setting block of size \" + (size / 1024.0)\n- + \" Kbytes to zero on \" + this);\n+ LOG.trace(\"GPU : in allocate from instruction \" + instructionName + \", setting block of size \" + (size\n+ / 1024.0) + \" Kbytes to zero on \" + this);\n}\ncudaMemset(A, 0, size);\nif (DMLScript.STATISTICS)\n@@ -395,12 +405,13 @@ public class GPUContext {\nreturn;\nlong t0 = 0;\nif (!cudaBlockSizeMap.containsKey(toFree))\n- throw new RuntimeException(\"ERROR : Internal state corrupted, cache block size map is not aware of a block it trying to free up\");\n+ throw new RuntimeException(\n+ \"ERROR : Internal state corrupted, cache block size map is not aware of a block it trying to free up\");\nlong size = cudaBlockSizeMap.get(toFree);\nif (eager) {\nif (LOG.isTraceEnabled()) {\n- LOG.trace(\"GPU : eagerly freeing cuda memory [ \" + toFree + \" ] for instruction \" + instructionName + \" on \"\n- + this);\n+ LOG.trace(\"GPU : eagerly freeing cuda memory [ \" + toFree + \" ] for instruction \" + instructionName\n+ + \" on \" + this);\n}\nif (DMLScript.STATISTICS)\nt0 = System.nanoTime();\n@@ -417,9 +428,9 @@ public class GPUContext {\nif (LOG.isTraceEnabled()) {\nLOG.trace(\"GPU : lazily freeing cuda memory for instruction \" + instructionName + \" on \" + this);\n}\n- LinkedList<Pointer> freeList = freeCUDASpaceMap.get(size);\n+ Set<Pointer> freeList = freeCUDASpaceMap.get(size);\nif (freeList == null) {\n- freeList = new LinkedList<Pointer>();\n+ freeList = new HashSet<>();\nfreeCUDASpaceMap.put(size, freeList);\n}\nif (freeList.contains(toFree))\n@@ -483,14 +494,18 @@ public class GPUContext {\nGPUStatistics.cudaEvictionCount.add(1);\n// Release the set of free blocks maintained in a GPUObject.freeCUDASpaceMap\n// to free up space\n- LRUCacheMap<Long, LinkedList<Pointer>> lruCacheMap = freeCUDASpaceMap;\n+ LRUCacheMap<Long, Set<Pointer>> lruCacheMap = freeCUDASpaceMap;\nwhile (lruCacheMap.size() > 0) {\nif (neededSize <= getAvailableMemory())\nbreak;\n- Map.Entry<Long, LinkedList<Pointer>> toFreeListPair = lruCacheMap.removeAndGetLRUEntry();\n- LinkedList<Pointer> toFreeList = toFreeListPair.getValue();\n+ Map.Entry<Long, Set<Pointer>> toFreeListPair = lruCacheMap.removeAndGetLRUEntry();\n+ Set<Pointer> toFreeList = toFreeListPair.getValue();\nLong size = toFreeListPair.getKey();\n- Pointer toFree = toFreeList.pop();\n+\n+ Iterator<Pointer> it = toFreeList.iterator(); // at this point, freeList should have at least one element\n+ Pointer toFree = it.next();\n+ it.remove();\n+\nif (toFreeList.isEmpty())\nlruCacheMap.remove(size);\ncudaFreeHelper(instructionName, toFree, true);\n@@ -548,7 +563,8 @@ public class GPUContext {\nGPUObject toBeRemoved = allocatedGPUObjects.get(allocatedGPUObjects.size() - 1);\nif (toBeRemoved.isLocked()) {\nthrow new DMLRuntimeException(\n- \"There is not enough memory on device for this matrix, request (\" + neededSize + \"). Allocated GPU objects:\" + allocatedGPUObjects.toString());\n+ \"There is not enough memory on device for this matrix, request (\" + neededSize\n+ + \"). Allocated GPU objects:\" + allocatedGPUObjects.toString());\n}\nif (toBeRemoved.dirty) {\ntoBeRemoved.copyFromDeviceToHost();\n@@ -697,7 +713,7 @@ public class GPUContext {\n* @return cudnnHandle for current thread\n*/\npublic cudnnHandle getCudnnHandle() {\n- return cudnnHandle.get();\n+ return cudnnHandle;\n}\n/**\n@@ -706,7 +722,7 @@ public class GPUContext {\n* @return cublasHandle for current thread\n*/\npublic cublasHandle getCublasHandle() {\n- return cublasHandle.get();\n+ return cublasHandle;\n}\n/**\n@@ -715,7 +731,7 @@ public class GPUContext {\n* @return cusparseHandle for current thread\n*/\npublic cusparseHandle getCusparseHandle() {\n- return cusparseHandle.get();\n+ return cusparseHandle;\n}\n/**\n@@ -724,7 +740,7 @@ public class GPUContext {\n* @return cusolverDnHandle for current thread\n*/\npublic cusolverDnHandle getCusolverDnHandle() {\n- return cusolverDnHandle.get();\n+ return cusolverDnHandle;\n}\n/**\n@@ -733,7 +749,7 @@ public class GPUContext {\n* @return cusolverSpHandle for current thread\n*/\npublic cusolverSpHandle getCusolverSpHandle() {\n- return cusolverSpHandle.get();\n+ return cusolverSpHandle;\n}\n/**\n@@ -742,7 +758,7 @@ public class GPUContext {\n* @return {@link JCudaKernels} for current thread\n*/\npublic JCudaKernels getKernels() {\n- return kernels.get();\n+ return kernels;\n}\n/**\n@@ -755,11 +771,34 @@ public class GPUContext {\nLOG.trace(\"GPU : this context was destroyed, this = \" + this.toString());\n}\nclearMemory();\n- cudnnDestroy(cudnnHandle.get());\n- cublasDestroy(cublasHandle.get());\n- cusparseDestroy(cusparseHandle.get());\n- cusolverDnDestroy(cusolverDnHandle.get());\n- cusolverSpDestroy(cusolverSpHandle.get());\n+\n+ deleteCudaLibraryHandles();\n+ }\n+\n+ /**\n+ * Deletes CUDA library handles\n+ */\n+ private void deleteCudaLibraryHandles() {\n+ if (cudnnHandle != null)\n+ cudnnDestroy(cudnnHandle);\n+\n+ if (cublasHandle != null)\n+ cublasDestroy(cublasHandle);\n+\n+ if (cusparseHandle != null)\n+ cusparseDestroy(cusparseHandle);\n+\n+ if (cusolverDnHandle != null)\n+ cusolverDnDestroy(cusolverDnHandle);\n+\n+ if (cusolverSpHandle != null)\n+ cusolverSpDestroy(cusolverSpHandle);\n+\n+ cudnnHandle = null;\n+ cublasHandle = null;\n+ cusparseHandle = null;\n+ cusolverDnHandle = null;\n+ cusolverSpHandle = null;\n}\n/**\n@@ -817,7 +856,7 @@ public class GPUContext {\n}\n// garbage collect all temporarily allocated spaces\n- for (LinkedList<Pointer> l : freeCUDASpaceMap.values()) {\n+ for (Set<Pointer> l : freeCUDASpaceMap.values()) {\nfor (Pointer p : l) {\ncudaFreeHelper(p, true);\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "<plugin>\n<groupId>com.github.os72</groupId>\n<artifactId>protoc-jar-maven-plugin</artifactId>\n- <version>3.4.0.1-SNAPSHOT</version>\n+ <version>3.0.0-b2.1</version>\n<executions>\n<execution>\n<id>caffe-sources</id>\n<goal>run</goal>\n</goals>\n<configuration>\n- <protocVersion>3.4.0</protocVersion> <!-- 2.4.1, 2.5.0, 2.6.1, 3.0.0 -->\n+ <protocVersion>3.0.0</protocVersion> <!-- 2.4.1, 2.5.0, 2.6.1, 3.0.0 -->\n<inputDirectories>\n<include>src/main/proto/tensorflow</include>\n</inputDirectories>\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/perftest/python/utils_misc.py", "new_path": "scripts/perftest/python/utils_misc.py", "diff": "@@ -72,8 +72,6 @@ def split_config_args(args):\nif 'config' in args.keys():\nif args['config'] is not None:\nsystemml_args_dict['-config'] = args['config']\n- else:\n- systemml_args_dict['-config'] = ''\nif 'gpu' in args.keys():\nif args['gpu'] is not None:\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/AutomatedTestBase.java", "new_path": "src/test/java/org/apache/sysml/test/integration/AutomatedTestBase.java", "diff": "@@ -874,12 +874,12 @@ public abstract class AutomatedTestBase\n// Create a SystemML config file for this test case based on default template\n// from src/test/config or derive from custom configuration provided by test.\nString configTemplate = FileUtils.readFileToString(getConfigTemplateFile(), \"UTF-8\");\n-\nString localTemp = curLocalTempDir.getPath();\n- String configContents = configTemplate.replace(\"<scratch>scratch_space</scratch>\",\n- String.format(\"<scratch>%s/scratch_space</scratch>\", localTemp));\n- configContents = configContents.replace(\"<localtmpdir>/tmp/systemml</localtmpdir>\",\n- String.format(\"<localtmpdir>%s/localtmp</localtmpdir>\", localTemp));\n+ String configContents = configTemplate\n+ .replace(createXMLElement(DMLConfig.SCRATCH_SPACE, \"scratch_space\"),\n+ createXMLElement(DMLConfig.SCRATCH_SPACE, localTemp+\"/scratch_space\"))\n+ .replace(createXMLElement(DMLConfig.LOCAL_TMP_DIR, \"/tmp/systemml\"),\n+ createXMLElement(DMLConfig.LOCAL_TMP_DIR, localTemp+\"/localtmp\"));\nFileUtils.write(getCurConfigFile(), configContents, \"UTF-8\");\n@@ -892,7 +892,9 @@ public abstract class AutomatedTestBase\nTestUtils.clearDirectory(DEBUG_TEMP_DIR + baseDirectory + INPUT_DIR);\n}\n-\n+ public String createXMLElement(String tagName, String value) {\n+ return String.format(\"<%s>%s</%s>\",tagName, value, tagName);\n+ }\n/**\n* <p>\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/datagen/genRandData4PCA.dml", "new_path": "scripts/datagen/genRandData4PCA.dml", "diff": "#\n#-------------------------------------------------------------\n-/*\n-Synthetic data generator for PCA.\n--> 3 hidden dimensions (V1, V2, V3)\n--> generates only \"dense\" data\n-\n----------------------------------\n- Parameters\n----------------------------------\n-$R = #rows\n-$C = #columns\n-$OUT = output file path on HDFS\n-$FMT = output format\n----------------------------------\n-hadoop jar SystemML.jar -f genRandData4PCA.dml -nvargs R=1000000 C=1000 DATA=/user/biuser/pcaData.mtx FMT=csv\n----------------------------------\n-*/\n-\n-FMT = ifdef($FMT,\"binary\"); # default output format\n-\n-# number of categorical attributes.. numC <= C\n-R = $R;\n-C = $C;\n+#\n+# Synthetic data generator for PCA\n+# 3 hidden dimensions (V1, V2, V3)\n+# generates only \"dense\" data\n+#\n+# INPUT PARAMETERS:\n+# --------------------------------------------------------------------------------------------\n+# NAME TYPE DEFAULT MEANING\n+# --------------------------------------------------------------------------------------------\n+# R Int 10000 Number of rows\n+# C Int 1000 Number of categorical attributes\n+# OUT String --- Location (on HDFS) to store the generated dataset\n+# FMT String \"csv\" Matrix output format, usually \"text\", \"csv\" or \"binary\"\n+# --------------------------------------------------------------------------------------------\n+#\n+# Example:\n+# hadoop jar SystemML.jar -f genRandData4PCA.dml -nvargs R=1000000 C=1000 OUT=/user/biuser/pcaData.mtx FMT=csv\n+\n+R = ifdef ($R, 10000)\n+C = ifdef ($C, 1000)\n+FMT = ifdef ($FMT, \"csv\");\n# Modofied version of the procedure from Zou et.al., \"Sparse Principal Component Analysis\", 2006.\n" }, { "change_type": "ADD", "old_path": null, "new_path": "scripts/staging/PCA.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+# This script performs Principal Component Analysis (PCA) on the given input data.\n+#\n+# INPUT PARAMETERS:\n+# ---------------------------------------------------------------------------------------------\n+# NAME TYPE DEFAULT MEANING\n+# ---------------------------------------------------------------------------------------------\n+# INPUT String --- Location to read the matrix A of feature vectors\n+# K Int --- Indicates dimension of the new vector space constructed from eigen vectors\n+# CENTER Int 0 Indicates whether or not to center data\n+# SCALE Int 0 Indicates whether or not to scale data\n+# OFMT String --- Output data format\n+# PROJDATA Int 0 This argument indicates if the data should be projected or not\n+# MODEL String --- Location to already existing model: eigenvectors and eigenvalues\n+# OUTPUT String / Location to write output matrices (covariance matrix, new basis vectors,\n+# and data projected onto new basis vectors)\n+# hadoop jar SystemML.jar -f PCA.dml -nvargs INPUT=INPUT_DIR/pca-1000x1000\n+# OUTPUT=OUTPUT_DIR/pca-1000x1000-model PROJDATA=1 CENTER=1 SCALE=1\n+# ---------------------------------------------------------------------------------------------\n+\n+A = read($INPUT);\n+K = ifdef($K, ncol(A));\n+ofmt = ifdef($OFMT, \"CSV\");\n+projectData = ifdef($PROJDATA,0);\n+model = ifdef($MODEL,\"\");\n+center = ifdef($CENTER,0);\n+scale = ifdef($SCALE,0);\n+output = ifdef($OUTPUT,\"/\");\n+\n+# reuse existing model to project data\n+if (model != \"\") {\n+ evec_dominant = read(model+\"/dominant.eigen.vectors\");\n+ }else{\n+ model = output;\n+}\n+\n+PCA = function(matrix[double] A, integer K, string ofmt, integer projectData, string model, integer center, integer scale, string output)\n+ return(matrix[double] eval_dominant, matrix[double] evec_dominant) {\n+\n+ evec_dominant = matrix(0,cols=1,rows=1);\n+\n+ N = nrow(A);\n+ D = ncol(A);\n+\n+ # perform z-scoring (centering and scaling)\n+ if (center == 1) {\n+ cm = colMeans(A);\n+ A = A - cm;\n+ }\n+ if (scale == 1) {\n+ cvars = (colSums (A^2));\n+ if (center == 1){\n+ cm = colMeans(A);\n+ cvars = (cvars - N*(cm^2))/(N-1);\n+ }\n+ Azscored = (A)/sqrt(cvars);\n+ A = Azscored;\n+ }\n+\n+ # co-variance matrix\n+ mu = colSums(A)/N;\n+ C = (t(A) %*% A)/(N-1) - (N/(N-1))*t(mu) %*% mu;\n+\n+\n+ # compute eigen vectors and values\n+ [evalues, evectors] = eigen(C);\n+\n+ decreasing_Idx = order(target=evalues,by=1,decreasing=TRUE,index.return=TRUE);\n+ diagmat = table(seq(1,D),decreasing_Idx);\n+ # sorts eigenvalues by decreasing order\n+ evalues = diagmat %*% evalues;\n+ # sorts eigenvectors column-wise in the order of decreasing eigenvalues\n+ evectors = evectors %*% diagmat;\n+\n+\n+ # select K dominant eigen vectors\n+ nvec = ncol(evectors);\n+\n+ eval_dominant = evalues[1:K, 1];\n+ evec_dominant = evectors[,1:K];\n+\n+}\n+\n+[eval_dominant, evec_dominant] = PCA(A, K, ofmt, projectData, model, center, scale, output)\n+\n+# the square root of eigenvalues\n+eval_stdev_dominant = sqrt(eval_dominant);\n+write(eval_stdev_dominant, model+\"/dominant.eigen.standard.deviations\", format=ofmt);\n+write(eval_dominant, model+\"/dominant.eigen.values\", format=ofmt);\n+write(evec_dominant, model+\"/dominant.eigen.vectors\", format=ofmt);\n+\n+# Construct new data set by treating computed dominant eigenvectors as the basis vectors\n+if (projectData == 1 | model != \"\"){\n+ newA = A %*% evec_dominant;\n+ write(newA, output+\"/projected.data\", format=ofmt);\n+}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofCellwise.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofCellwise.java", "diff": "@@ -116,10 +116,6 @@ public abstract class SpoofCellwise extends SpoofOperator implements Serializabl\nif( inputs==null || inputs.size() < 1 )\nthrow new RuntimeException(\"Invalid input arguments.\");\n- if( getTotalInputNnz(inputs) < PAR_NUMCELL_THRESHOLD ) {\n- k = 1; //serial execution\n- }\n-\n//input preparation\nMatrixBlock a = inputs.get(0);\nSideInput[] b = prepInputMatrices(inputs);\n@@ -131,6 +127,12 @@ public abstract class SpoofCellwise extends SpoofOperator implements Serializabl\nboolean sparseSafe = isSparseSafe() || (b.length == 0\n&& genexec( 0, b, scalars, m, n, 0, 0 ) == 0);\n+ long inputSize = sparseSafe ?\n+ getTotalInputNnz(inputs) : getTotalInputSize(inputs);\n+ if( inputSize < PAR_NUMCELL_THRESHOLD ) {\n+ k = 1; //serial execution\n+ }\n+\ndouble ret = 0;\nif( k <= 1 ) //SINGLE-THREADED\n{\n@@ -199,10 +201,6 @@ public abstract class SpoofCellwise extends SpoofOperator implements Serializabl\nif( inputs==null || inputs.size() < 1 || out==null )\nthrow new RuntimeException(\"Invalid input arguments.\");\n- if( getTotalInputNnz(inputs) < PAR_NUMCELL_THRESHOLD ) {\n- k = 1; //serial execution\n- }\n-\n//input preparation\nMatrixBlock a = inputs.get(0);\nSideInput[] b = prepInputMatrices(inputs);\n@@ -214,6 +212,12 @@ public abstract class SpoofCellwise extends SpoofOperator implements Serializabl\nboolean sparseSafe = isSparseSafe() || (b.length == 0\n&& genexec( 0, b, scalars, m, n, 0, 0 ) == 0);\n+ long inputSize = sparseSafe ?\n+ getTotalInputNnz(inputs) : getTotalInputSize(inputs);\n+ if( inputSize < PAR_NUMCELL_THRESHOLD ) {\n+ k = 1; //serial execution\n+ }\n+\n//result allocation and preparations\nboolean sparseOut = _type == CellType.NO_AGG\n&& sparseSafe && a.isInSparseFormat();\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofMultiAggregate.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofMultiAggregate.java", "diff": "@@ -85,7 +85,9 @@ public abstract class SpoofMultiAggregate extends SpoofOperator implements Seria\nif( inputs==null || inputs.size() < 1 )\nthrow new RuntimeException(\"Invalid input arguments.\");\n- if( getTotalInputNnz(inputs) < PAR_NUMCELL_THRESHOLD ) {\n+ long inputSize = isSparseSafe() ?\n+ getTotalInputNnz(inputs) : getTotalInputSize(inputs);\n+ if( inputSize < PAR_NUMCELL_THRESHOLD ) {\nk = 1; //serial execution\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofOperator.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofOperator.java", "diff": "@@ -156,6 +156,11 @@ public abstract class SpoofOperator implements Serializable\nreturn inputs.stream().mapToLong(in -> in.getNonZeros()).sum();\n}\n+ public static long getTotalInputSize(ArrayList<MatrixBlock> inputs) {\n+ return inputs.stream().mapToLong(\n+ in -> (long)in.getNumRows() * in.getNumColumns()).sum();\n+ }\n+\n//abstraction for safely accessing sideways matrices without the need\n//to allocate empty matrices as dense, see prepInputMatrices\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofRowwise.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofRowwise.java", "diff": "@@ -170,7 +170,7 @@ public abstract class SpoofRowwise extends SpoofOperator\n{\n//redirect to serial execution\nif( k <= 1 || (_type.isColumnAgg() && !LibMatrixMult.checkParColumnAgg(inputs.get(0), k, false))\n- || getTotalInputNnz(inputs) < PAR_NUMCELL_THRESHOLD ) {\n+ || getTotalInputSize(inputs) < PAR_NUMCELL_THRESHOLD ) {\nreturn execute(inputs, scalarObjects, out);\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/GPUInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/GPUInstruction.java", "diff": "@@ -61,6 +61,7 @@ public abstract class GPUInstruction extends Instruction {\npublic final static String MISC_TIMER_ROW_TO_COLUMN_MAJOR = \"r2c\"; // time spent in converting data from row major to column major\npublic final static String MISC_TIMER_COLUMN_TO_ROW_MAJOR = \"c2r\"; // time spent in converting data from column major to row major\npublic final static String MISC_TIMER_OBJECT_CLONE = \"clone\";// time spent in cloning (deep copying) a GPUObject instance\n+ public final static String MISC_TIMER_CUDA_SYNC = \"sync\"; // time spent in device sync\npublic final static String MISC_TIMER_CUDA_FREE = \"f\"; // time spent in calling cudaFree\npublic final static String MISC_TIMER_ALLOCATE = \"a\"; // time spent to allocate memory on gpu\n@@ -198,7 +199,11 @@ public abstract class GPUInstruction extends Instruction {\nthrows DMLRuntimeException\n{\nif(DMLScript.SYNCHRONIZE_GPU) {\n+ long t0 = GPUStatistics.DISPLAY_STATISTICS ? System.nanoTime() : 0;\njcuda.runtime.JCuda.cudaDeviceSynchronize();\n+ if(GPUStatistics.DISPLAY_STATISTICS) {\n+ GPUStatistics.maintainCPMiscTimes(getExtendedOpcode(), GPUInstruction.MISC_TIMER_CUDA_SYNC, System.nanoTime() - t0);\n+ }\n}\nif(LOG.isDebugEnabled()) {\nfor(GPUContext gpuCtx : ec.getGPUContexts()) {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixCuDNN.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixCuDNN.java", "diff": "@@ -67,6 +67,7 @@ import jcuda.jcudnn.cudnnTensorDescriptor;\nimport org.apache.commons.logging.Log;\nimport org.apache.commons.logging.LogFactory;\n+import org.apache.sysml.api.DMLScript;\nimport org.apache.sysml.hops.OptimizerUtils;\nimport org.apache.sysml.runtime.DMLRuntimeException;\nimport org.apache.sysml.runtime.controlprogram.caching.MatrixObject;\n@@ -153,7 +154,8 @@ public class LibMatrixCuDNN extends LibMatrixCUDA {\nlong CHW = C*H*W; long KPQ = K*P*Q; long CRS = C*R*S;\nlong NCHW = N*CHW; long NKPQ = N*KPQ; long KCRS = K*CRS;\n- if(NCHW < maxNumDoublesOfCuDNNTensor && NKPQ < maxNumDoublesOfCuDNNTensor && KCRS < maxNumDoublesOfCuDNNTensor) {\n+ if(DMLScript.FORCE_ACCELERATOR ||\n+ (NCHW < maxNumDoublesOfCuDNNTensor && NKPQ < maxNumDoublesOfCuDNNTensor && KCRS < maxNumDoublesOfCuDNNTensor)) {\n// Filter and output are accounted as dense in the memory estimation for conv2d\ndouble overhead = isInSparseFormat(gCtx, filter) ? OptimizerUtils.estimateSizeExactSparsity(K, CRS, 1.0) : 0;\noverhead += isInSparseFormat(gCtx, image) ? OptimizerUtils.estimateSizeExactSparsity(N, CHW, 1.0) : 0;\n@@ -161,7 +163,7 @@ public class LibMatrixCuDNN extends LibMatrixCUDA {\nPointer filterPointer = getDensePointerForCuDNN(gCtx, filter, instName);\nPointer dstPointer = getDensePointerForCuDNN(gCtx, outputBlock, instName);\n- if(overhead <= intermediateMemoryBudget) {\n+ if(DMLScript.FORCE_ACCELERATOR || overhead <= intermediateMemoryBudget) {\n// Perform all-input all-channel conv2d\nPointer imagePointer = getDensePointerForCuDNN(gCtx, image, instName);\ncudnnConv2d(gCtx, instName, imagePointer, filterPointer, dstPointer, N, C, H, W, K, R, S, pad_h, pad_w, stride_h, stride_w, P, Q);\n@@ -346,11 +348,12 @@ public class LibMatrixCuDNN extends LibMatrixCUDA {\nlong CHW = C*H*W; long KPQ = K*P*Q; long CRS = C*R*S;\nlong NCHW = N*CHW; long NKPQ = N*KPQ; long KCRS = K*CRS;\n- if(NCHW < maxNumDoublesOfCuDNNTensor && NKPQ < maxNumDoublesOfCuDNNTensor && KCRS < maxNumDoublesOfCuDNNTensor) {\n+ if(DMLScript.FORCE_ACCELERATOR ||\n+ (NCHW < maxNumDoublesOfCuDNNTensor && NKPQ < maxNumDoublesOfCuDNNTensor && KCRS < maxNumDoublesOfCuDNNTensor)) {\nPointer dwPointer = getDensePointerForCuDNN(gCtx, outputBlock, instName);\ndouble overhead = isInSparseFormat(gCtx, image) ? OptimizerUtils.estimateSizeExactSparsity(N, CHW, 1.0) : 0;\noverhead += isInSparseFormat(gCtx, dout) ? OptimizerUtils.estimateSizeExactSparsity(N, KPQ, 1.0) : 0;\n- if(overhead <= intermediateMemoryBudget) {\n+ if(DMLScript.FORCE_ACCELERATOR || overhead <= intermediateMemoryBudget) {\n// Perform all-input all-channel conv2dBackwardFilter\nPointer imagePointer = getDensePointerForCuDNN(gCtx, image, instName);\nPointer doutPointer = getDensePointerForCuDNN(gCtx, dout, instName);\n@@ -502,13 +505,14 @@ public class LibMatrixCuDNN extends LibMatrixCUDA {\nlong CHW = C*H*W; long KPQ = K*P*Q; long CRS = C*R*S;\nlong NCHW = N*CHW; long NKPQ = N*KPQ; long KCRS = K*CRS;\n- if(NCHW < maxNumDoublesOfCuDNNTensor && NKPQ < maxNumDoublesOfCuDNNTensor && KCRS < maxNumDoublesOfCuDNNTensor) {\n+ if(DMLScript.FORCE_ACCELERATOR ||\n+ (NCHW < maxNumDoublesOfCuDNNTensor && NKPQ < maxNumDoublesOfCuDNNTensor && KCRS < maxNumDoublesOfCuDNNTensor)) {\n// Filter and output are accounted as dense in the memory estimation for conv2dBackwardData\ndouble overhead = isInSparseFormat(gCtx, filter) ? OptimizerUtils.estimateSizeExactSparsity(K, CRS, 1.0) : 0;\noverhead += isInSparseFormat(gCtx, dout) ? OptimizerUtils.estimateSizeExactSparsity(N, KPQ, 1.0) : 0;\nPointer filterPointer = getDensePointerForCuDNN(gCtx, filter, instName);\nPointer dstPointer = getDensePointerForCuDNN(gCtx, output, instName);\n- if(overhead <= intermediateMemoryBudget) {\n+ if(DMLScript.FORCE_ACCELERATOR || overhead <= intermediateMemoryBudget) {\n// Perform all-input all-channel conv2dBackwardData\nPointer doutPointer = getDensePointerForCuDNN(gCtx, dout, instName);\ncudnnConv2dBackwardData(gCtx, instName, filterPointer, doutPointer, dstPointer,\n@@ -638,11 +642,12 @@ public class LibMatrixCuDNN extends LibMatrixCUDA {\nlong CHW = C*H*W; long CPQ = C*P*Q;\nlong NCHW = N*CHW; long NCPQ = N*CPQ;\n- if(NCHW < maxNumDoublesOfCuDNNTensor && NCPQ < maxNumDoublesOfCuDNNTensor) {\n+ if(DMLScript.FORCE_ACCELERATOR ||\n+ (NCHW < maxNumDoublesOfCuDNNTensor && NCPQ < maxNumDoublesOfCuDNNTensor)) {\n// Filter and output are accounted as dense in the memory estimation for conv2dBackwardData\nlong overhead = isInSparseFormat(gCtx, image) ? OptimizerUtils.estimateSizeExactSparsity(N, CHW, 1.0) : 0;\nPointer y = getDensePointerForCuDNN(gCtx, outputBlock, instName);\n- if(overhead <= intermediateMemoryBudget) {\n+ if(DMLScript.FORCE_ACCELERATOR || overhead <= intermediateMemoryBudget) {\nPointer x = getDensePointerForCuDNN(gCtx, image, instName);\ncudnnTensorDescriptor xDesc = allocateTensorDescriptor(gCtx, image, N, C, H, W);\ncudnnMaxpooling(gCtx, instName, x, xDesc, y, N, C, H, W, K, R, S, pad_h, pad_w, stride_h, stride_w, P, Q);\n@@ -780,12 +785,13 @@ public class LibMatrixCuDNN extends LibMatrixCUDA {\nlong CHW = C*H*W; long CPQ = C*P*Q;\nlong NCHW = N*CHW; long NCPQ = N*CPQ;\n- if(NCHW < maxNumDoublesOfCuDNNTensor && NCPQ < maxNumDoublesOfCuDNNTensor) {\n+ if(DMLScript.FORCE_ACCELERATOR ||\n+ (NCHW < maxNumDoublesOfCuDNNTensor && NCPQ < maxNumDoublesOfCuDNNTensor)) {\n// Filter and output are accounted as dense in the memory estimation for conv2dBackwardData\nlong overhead = isInSparseFormat(gCtx, image) ? OptimizerUtils.estimateSizeExactSparsity(N, CHW, 1.0) : 0;\noverhead += isInSparseFormat(gCtx, dout) ? OptimizerUtils.estimateSizeExactSparsity(N, CPQ, 1.0) : 0;\nPointer dx = getDensePointerForCuDNN(gCtx, outputBlock, instName);\n- if(overhead <= intermediateMemoryBudget) {\n+ if(DMLScript.FORCE_ACCELERATOR || overhead <= intermediateMemoryBudget) {\nPointer x = getDensePointerForCuDNN(gCtx, image, instName);\nPointer dy = getDensePointerForCuDNN(gCtx, dout, instName);\ncudnnMaxpoolingBackward(gCtx, instName, x, dy, dx, N, C, H, W, K, R, S, pad_h, pad_w, stride_h, stride_w, P, Q);\n" } ]

[ { "change_type": "MODIFY", "old_path": "bin/systemml-spark-submit.py", "new_path": "bin/systemml-spark-submit.py", "diff": "@@ -92,6 +92,7 @@ def spark_submit_entry(master, driver_memory, num_executors, executor_memory,\nml_options.append(stats)\nif gpu is not None:\nml_options.append('-gpu')\n+ if gpu is not 'no_option':\nml_options.append(gpu)\nif len(ml_options) < 1:\n@@ -99,18 +100,27 @@ def spark_submit_entry(master, driver_memory, num_executors, executor_memory,\n# stats, explain, target_jars\ncmd_spark = [spark_path, '--class', 'org.apache.sysml.api.DMLScript',\n- '--master', master, '--driver-memory', driver_memory,\n- '--num-executors', num_executors, '--executor-memory', executor_memory,\n- '--executor-cores', executor_cores, '--conf', default_conf,\n+ '--master', master,\n+ '--driver-memory', driver_memory,\n+ '--conf', default_conf,\n'--jars', cuda_jars, systemml_jars]\n+ if num_executors is not None:\n+ cmd_spark = cmd_spark + ['--num-executors', num_executors]\n+\n+ if executor_memory is not None:\n+ cmd_spark = cmd_spark + ['--executor-memory', executor_memory]\n+\n+ if executor_cores is not None:\n+ cmd_spark = cmd_spark + ['--executor-cores', executor_cores]\n+\ncmd_system_ml = ['-config', default_config,\n'-exec', 'hybrid_spark', '-f', script_file, ' '.join(ml_options)]\ncmd = cmd_spark + cmd_system_ml\n# Debug\n- # print(' '.join(cmd))\n+ print(' '.join(cmd))\nreturn_code = os.system(' '.join(cmd))\nreturn return_code\n@@ -120,10 +130,10 @@ if __name__ == '__main__':\ndescription='System-ML Spark Submit Script')\n# SPARK-SUBMIT Options\ncparser.add_argument('--master', default='local[*]', help='local, yarn-client, yarn-cluster', metavar='')\n- cparser.add_argument('--driver-memory', default='5G', help='Memory for driver (e.g. 512M)', metavar='')\n- cparser.add_argument('--num-executors', default='2', help='Number of executors to launch', metavar='')\n- cparser.add_argument('--executor-memory', default='2G', help='Memory per executor', metavar='')\n- cparser.add_argument('--executor-cores', default='1', help='Number of cores', metavar='')\n+ cparser.add_argument('--driver-memory', default='8G', help='Memory for driver (e.g. 512M, 1G)', metavar='')\n+ cparser.add_argument('--num-executors', nargs=1, help='Number of executors to launch', metavar='')\n+ cparser.add_argument('--executor-memory', nargs=1, help='Memory per executor', metavar='')\n+ cparser.add_argument('--executor-cores', nargs=1, help='Number of executor cores', metavar='')\ncparser.add_argument('--conf', help='Spark configuration file', nargs='+', metavar='')\n# SYSTEM-ML Options\n@@ -138,7 +148,7 @@ if __name__ == '__main__':\nmetavar='')\ncparser.add_argument('-gpu', help='uses CUDA instructions when reasonable, '\n'set <force> option to skip conservative memory estimates '\n- 'and use GPU wherever possible', nargs='?')\n+ 'and use GPU wherever possible', nargs='?', const='no_option')\ncparser.add_argument('-f', required=True, help='specifies dml/pydml file to execute; '\n'path can be local/hdfs/gpfs', metavar='')\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/datagen.py", "new_path": "scripts/perftest/python/datagen.py", "diff": "@@ -25,7 +25,7 @@ from os.path import join\nfrom utils_misc import split_rowcol, config_writer, mat_type_check\n# This file contains configuration settings for data generation\n-DATA_FORMAT = 'csv'\n+DATA_FORMAT = 'binary'\nMATRIX_TYPE_DICT = {'dense': '0.9',\n'sparse': '0.01'}\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/predict.py", "new_path": "scripts/perftest/python/predict.py", "diff": "@@ -26,7 +26,7 @@ from utils_misc import config_writer, mat_type_check\nfrom utils_fs import relevant_folders\n# Contains configuration setting for predicting\n-DATA_FORMAT = 'csv'\n+DATA_FORMAT = 'binary'\ndef m_svm_predict(save_folder_name, datagen_dir, train_dir, predict_dir, config_dir):\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/run_perftest.py", "new_path": "scripts/perftest/python/run_perftest.py", "diff": "@@ -141,7 +141,7 @@ def algorithm_workflow(algo, exec_type, config_path, dml_file_name, action_mode,\nreturn exit_flag_success\n-def perf_test_entry(family, algo, exec_type, mat_type, mat_shape, config_dir, mode, temp_dir):\n+def perf_test_entry(family, algo, exec_type, mat_type, mat_shape, config_dir, mode, temp_dir, file_system_type):\n\"\"\"\nThis function is the entry point for performance testing\n@@ -168,6 +168,9 @@ def perf_test_entry(family, algo, exec_type, mat_type, mat_shape, config_dir, mo\ntemp_dir: String\nLocation to store all output files created during perf test\n+\n+ file_system_type: String\n+\n\"\"\"\n# algos to run is a list of tuples with\n# [(m-svm, binomial), (m-svm, multinomial)...]\n@@ -275,6 +278,7 @@ if __name__ == '__main__':\nmat_type = ['dense', 'sparse', 'all']\nworkload = ['data-gen', 'train', 'predict']\nexecution_mode = ['hybrid_spark', 'singlenode']\n+ file_system_type = ['hdfs', 'local']\n# Default Arguments\ndefault_mat_shape = ['10k_100']\n@@ -308,7 +312,6 @@ if __name__ == '__main__':\n'spark.driver.extraJavaOptions=\\\"-Xms20g -Xmn2g\\\"'\n-\n# Argparse Module\ncparser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter,\ndescription='SystemML Performance Test Script')\n@@ -335,8 +338,12 @@ if __name__ == '__main__':\ncparser.add_argument('--mode', default=workload,\nhelp='space separated list of types of workloads to run (available: data-gen, train, predict)',\nmetavar='', choices=workload, nargs='+')\n- # Change this to temp-dir\n- cparser.add_argument('--temp-dir', help='define the file system to work on', metavar='')\n+ cparser.add_argument('--temp-dir', help='the path on the file system to place the working temporary directory at',\n+ metavar='')\n+ cparser.add_argument('--file-system-type', choices=file_system_type, metavar='',\n+ help='file system for temp directory, '\n+ 'supported types are \\'hdfs\\' for hybrid_spark and \\'local\\' for standalone;'\n+ 'default for hybrid_spark is \\'hdfs\\' and for standalone is \\'local\\'')\n# Configuration Options\ncparser.add_argument('-stats', help='Monitor and report caching/recompilation statistics, '\n@@ -347,7 +354,7 @@ if __name__ == '__main__':\ncparser.add_argument('-config', help='System-ML configuration file (e.g SystemML-config.xml)', metavar='')\ncparser.add_argument('-gpu', help='uses CUDA instructions when reasonable, '\n'set <force> option to skip conservative memory estimates '\n- 'and use GPU wherever possible', nargs='?')\n+ 'and use GPU wherever possible', nargs='?', const='no_option')\n# Spark Configuration Option\ncparser.add_argument('--master', help='local, yarn-client, yarn-cluster', metavar='')\ncparser.add_argument('--driver-memory', help='Memory for driver (e.g. 512M)', metavar='')\n@@ -371,7 +378,7 @@ if __name__ == '__main__':\nperftest_args_dict, systemml_args_dict, backend_args_dict = split_config_args(all_arg_dict)\n# temp_dir hdfs / local path check\n- perftest_args_dict['temp_dir'] = get_default_dir(args.temp_dir, args.exec_type, default_config_dir)\n+ perftest_args_dict['temp_dir'] = get_default_dir(args.file_system_type, args.temp_dir, args.exec_type, default_config_dir)\n# default_mat_type validity\nif len(args.mat_type) > 2:\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/train.py", "new_path": "scripts/perftest/python/train.py", "diff": "@@ -27,7 +27,7 @@ from functools import reduce\nfrom utils_fs import relevant_folders\n# Contains configuration setting for training\n-DATA_FORMAT = 'csv'\n+DATA_FORMAT = 'binary'\ndef binomial_m_svm_train(save_folder_name, datagen_dir, train_dir, config_dir):\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/utils_misc.py", "new_path": "scripts/perftest/python/utils_misc.py", "diff": "@@ -54,30 +54,28 @@ def split_config_args(args):\nperftest_args_dict['filename'] = args['filename']\nperftest_args_dict['mode'] = args['mode']\nperftest_args_dict['temp_dir'] = args['temp_dir']\n+ perftest_args_dict['file_system_type'] = args['file_system_type']\nsystemml_args_dict = {}\n- if 'stats' in args.keys():\nif args['stats'] is not None:\nsystemml_args_dict['-stats'] = args['stats']\nelse:\nsystemml_args_dict['-stats'] = ''\n- if 'explain' in args.keys():\nif args['explain'] is not None:\nsystemml_args_dict['-explain'] = args['explain']\nelse:\nsystemml_args_dict['-explain'] = ''\n- if 'config' in args.keys():\nif args['config'] is not None:\nsystemml_args_dict['-config'] = args['config']\n- if 'gpu' in args.keys():\nif args['gpu'] is not None:\n- systemml_args_dict['-gpu'] = args['gpu']\n- else:\n+ if args['gpu'] == 'no_option':\nsystemml_args_dict['-gpu'] = ''\n+ else:\n+ systemml_args_dict['-gpu'] = args['gpu']\nbackend_args_dict = {}\nexec_type = args['exec_type']\n@@ -373,8 +371,9 @@ def mat_type_check(current_family, matrix_types, dense_algos):\nreturn current_type\n-def get_default_dir(temp_dir, exec_mode, config_dir):\n+def get_default_dir(file_system_type, temp_dir, exec_mode, config_dir):\n\"\"\"\n+ file_system_type: String\ntemp_dir: String\nexec_mode: String\nconfig_dir: String\n@@ -390,6 +389,7 @@ def get_default_dir(temp_dir, exec_mode, config_dir):\nreturn temp_dir\nif exec_mode == 'hybrid_spark':\n+ if file_system_type == 'hdfs':\ncmd = ['hdfs', 'getconf', '-confKey', 'fs.default.name']\nhdfs_base = subprocess_exec(' '.join(cmd), extract='hdfs_base')\n@@ -404,3 +404,8 @@ def get_default_dir(temp_dir, exec_mode, config_dir):\nelse:\nhdfs_home = join(hdfs_base, 'user', getpass.getuser(), temp_dir)\nreturn hdfs_home\n+ else:\n+ if temp_dir is None:\n+ return config_dir\n+ if temp_dir is not None:\n+ return temp_dir\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/perftest/python/run_perftest.py", "new_path": "scripts/perftest/python/run_perftest.py", "diff": "@@ -378,6 +378,12 @@ if __name__ == '__main__':\nperftest_args_dict, systemml_args_dict, backend_args_dict = split_config_args(all_arg_dict)\n# temp_dir hdfs / local path check\n+ if args.file_system_type is None:\n+ if args.exec_type == 'hybrid_spark':\n+ args.file_system_type = 'hdfs'\n+ else:\n+ args.file_system_type = 'local'\n+\nperftest_args_dict['temp_dir'] = get_default_dir(args.file_system_type, args.temp_dir, args.exec_type, default_config_dir)\n# default_mat_type validity\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "<plugin>\n<groupId>com.github.os72</groupId>\n<artifactId>protoc-jar-maven-plugin</artifactId>\n- <version>3.0.0-b2.1</version>\n+ <version>3.4.0.1</version>\n<executions>\n<execution>\n<id>caffe-sources</id>\n<goal>run</goal>\n</goals>\n<configuration>\n- <protocVersion>2.5.0</protocVersion> <!-- 2.4.1, 2.5.0, 2.6.1, 3.0.0 -->\n+ <protocVersion>2.6.1</protocVersion> <!-- 2.4.1, 2.5.0, 2.6.1, 3.4.0 -->\n<inputDirectories>\n<include>src/main/proto/caffe</include>\n</inputDirectories>\n" } ]

[ { "change_type": "MODIFY", "old_path": "conf/SystemML-config.xml.template", "new_path": "conf/SystemML-config.xml.template", "diff": "<!-- set the codegen java compiler (auto, janino, javac) -->\n<sysml.codegen.compiler>auto</sysml.codegen.compiler>\n+ <!-- set the codegen optimizer (fuse_all, fuse_no_redundancy, fuse_cost_based_v2) -->\n+ <sysml.codegen.compiler>fuse_cost_based_v2</sysml.codegen.compiler>\n+\n<!-- if codegen.enabled, enables source code caching of fused operators -->\n<sysml.codegen.plancache>true</sysml.codegen.plancache>\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/conf/DMLConfig.java", "new_path": "src/main/java/org/apache/sysml/conf/DMLConfig.java", "diff": "@@ -41,6 +41,7 @@ import org.apache.hadoop.fs.FileSystem;\nimport org.apache.hadoop.fs.Path;\nimport org.apache.sysml.hops.OptimizerUtils;\nimport org.apache.sysml.hops.codegen.SpoofCompiler.CompilerType;\n+import org.apache.sysml.hops.codegen.SpoofCompiler.PlanSelector;\nimport org.apache.sysml.lops.Compression;\nimport org.apache.sysml.parser.ParseException;\nimport org.apache.sysml.runtime.DMLRuntimeException;\n@@ -77,6 +78,7 @@ public class DMLConfig\npublic static final String NATIVE_BLAS = \"sysml.native.blas\";\npublic static final String CODEGEN = \"sysml.codegen.enabled\"; //boolean\npublic static final String CODEGEN_COMPILER = \"sysml.codegen.compiler\"; //see SpoofCompiler.CompilerType\n+ public static final String CODEGEN_OPTIMIZER = \"sysml.codegen.optimizer\"; //see SpoofCompiler.PlanSelector\npublic static final String CODEGEN_PLANCACHE = \"sysml.codegen.plancache\"; //boolean\npublic static final String CODEGEN_LITERALS = \"sysml.codegen.literals\"; //1..heuristic, 2..always\n@@ -125,6 +127,7 @@ public class DMLConfig\n_defaultVals.put(COMPRESSED_LINALG, Compression.CompressConfig.AUTO.name() );\n_defaultVals.put(CODEGEN, \"false\" );\n_defaultVals.put(CODEGEN_COMPILER, CompilerType.AUTO.name() );\n+ _defaultVals.put(CODEGEN_COMPILER, PlanSelector.FUSE_COST_BASED_V2.name() );\n_defaultVals.put(CODEGEN_PLANCACHE, \"true\" );\n_defaultVals.put(CODEGEN_LITERALS, \"1\" );\n_defaultVals.put(NATIVE_BLAS, \"none\" );\n@@ -416,7 +419,7 @@ public class DMLConfig\nYARN_APPMASTER, YARN_APPMASTERMEM, YARN_MAPREDUCEMEM,\nCP_PARALLEL_OPS, CP_PARALLEL_IO, NATIVE_BLAS,\nCOMPRESSED_LINALG,\n- CODEGEN, CODEGEN_COMPILER, CODEGEN_PLANCACHE, CODEGEN_LITERALS,\n+ CODEGEN, CODEGEN_COMPILER, CODEGEN_OPTIMIZER, CODEGEN_PLANCACHE, CODEGEN_LITERALS,\nEXTRA_GPU_STATS, EXTRA_DNN_STATS, EXTRA_FINEGRAINED_STATS, STATS_MAX_WRAP_LEN,\nAVAILABLE_GPUS, SYNCHRONIZE_GPU, EAGER_CUDA_FREE\n};\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/SpoofCompiler.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/SpoofCompiler.java", "diff": "@@ -108,12 +108,12 @@ public class SpoofCompiler\n//internal configuration flags\npublic static boolean LDEBUG = false;\npublic static CompilerType JAVA_COMPILER = CompilerType.JANINO;\n+ public static PlanSelector PLAN_SEL_POLICY = PlanSelector.FUSE_COST_BASED_V2;\npublic static IntegrationType INTEGRATION = IntegrationType.RUNTIME;\npublic static final boolean RECOMPILE_CODEGEN = true;\npublic static final boolean PRUNE_REDUNDANT_PLANS = true;\npublic static PlanCachePolicy PLAN_CACHE_POLICY = PlanCachePolicy.CSLH;\npublic static final int PLAN_CACHE_SIZE = 1024; //max 1K classes\n- public static final PlanSelector PLAN_SEL_POLICY = PlanSelector.FUSE_COST_BASED_V2;\npublic enum CompilerType {\nAUTO,\n@@ -486,6 +486,13 @@ public class SpoofCompiler\n}\n}\n+ public static void setConfiguredPlanSelector() {\n+ DMLConfig conf = ConfigurationManager.getDMLConfig();\n+ String optimizer = conf.getTextValue(DMLConfig.CODEGEN_OPTIMIZER);\n+ PlanSelector type = PlanSelector.valueOf(optimizer.toUpperCase());\n+ PLAN_SEL_POLICY = type;\n+ }\n+\npublic static void setExecTypeSpecificJavaCompiler() {\nDMLConfig conf = ConfigurationManager.getDMLConfig();\nString compiler = conf.getTextValue(DMLConfig.CODEGEN_COMPILER);\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/DMLTranslator.java", "new_path": "src/main/java/org/apache/sysml/parser/DMLTranslator.java", "diff": "@@ -297,6 +297,7 @@ public class DMLTranslator\nSpoofCompiler.PLAN_CACHE_POLICY = PlanCachePolicy.get(\ndmlconf.getBooleanValue(DMLConfig.CODEGEN_PLANCACHE),\ndmlconf.getIntValue(DMLConfig.CODEGEN_LITERALS)==2);\n+ SpoofCompiler.setConfiguredPlanSelector();\nSpoofCompiler.setExecTypeSpecificJavaCompiler();\nif( SpoofCompiler.INTEGRATION==IntegrationType.HOPS )\ncodgenHopsDAG(dmlp);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/conf/DMLConfig.java", "new_path": "src/main/java/org/apache/sysml/conf/DMLConfig.java", "diff": "@@ -127,7 +127,7 @@ public class DMLConfig\n_defaultVals.put(COMPRESSED_LINALG, Compression.CompressConfig.AUTO.name() );\n_defaultVals.put(CODEGEN, \"false\" );\n_defaultVals.put(CODEGEN_COMPILER, CompilerType.AUTO.name() );\n- _defaultVals.put(CODEGEN_COMPILER, PlanSelector.FUSE_COST_BASED_V2.name() );\n+ _defaultVals.put(CODEGEN_OPTIMIZER, PlanSelector.FUSE_COST_BASED_V2.name() );\n_defaultVals.put(CODEGEN_PLANCACHE, \"true\" );\n_defaultVals.put(CODEGEN_LITERALS, \"1\" );\n_defaultVals.put(NATIVE_BLAS, \"none\" );\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/SpoofFusedOp.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/SpoofFusedOp.java", "diff": "@@ -42,6 +42,7 @@ public class SpoofFusedOp extends Hop implements MultiThreadedHop\nROW_DIMS,\nCOLUMN_DIMS_ROWS,\nCOLUMN_DIMS_COLS,\n+ RANK_DIMS_COLS,\nSCALAR,\nMULTI_SCALAR,\nROW_RANK_DIMS, // right wdivmm, row mm\n@@ -163,6 +164,12 @@ public class SpoofFusedOp extends Hop implements MultiThreadedHop\ncase COLUMN_DIMS_COLS:\nret = new long[]{1, mc.getCols(), -1};\nbreak;\n+ case RANK_DIMS_COLS: {\n+ MatrixCharacteristics mc2 = memo.getAllInputStats(getInput().get(1));\n+ if( mc2.dimsKnown() )\n+ ret = new long[]{1, mc2.getCols(), -1};\n+ break;\n+ }\ncase INPUT_DIMS:\nret = new long[]{mc.getRows(), mc.getCols(), -1};\nbreak;\n@@ -219,6 +226,10 @@ public class SpoofFusedOp extends Hop implements MultiThreadedHop\nsetDim1(1);\nsetDim2(getInput().get(0).getDim2());\nbreak;\n+ case RANK_DIMS_COLS:\n+ setDim1(1);\n+ setDim2(getInput().get(1).getDim2());\n+ break;\ncase INPUT_DIMS:\nsetDim1(getInput().get(0).getDim1());\nsetDim2(getInput().get(0).getDim2());\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeRow.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeRow.java", "diff": "@@ -159,6 +159,7 @@ public class CNodeRow extends CNodeTpl\ncase COL_AGG_T: return SpoofOutputDimsType.COLUMN_DIMS_ROWS; //column vector\ncase COL_AGG_B1: return SpoofOutputDimsType.COLUMN_RANK_DIMS;\ncase COL_AGG_B1_T: return SpoofOutputDimsType.COLUMN_RANK_DIMS_T;\n+ case COL_AGG_B1R: return SpoofOutputDimsType.RANK_DIMS_COLS;\ndefault:\nthrow new RuntimeException(\"Unsupported row type: \"+_type.toString());\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/opt/PlanSelection.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/opt/PlanSelection.java", "diff": "@@ -34,6 +34,9 @@ import org.apache.sysml.runtime.util.UtilFunctions;\npublic abstract class PlanSelection\n{\n+ private static final BasicPlanComparator BASE_COMPARE = new BasicPlanComparator();\n+ private final TypedPlanComparator _typedCompare = new TypedPlanComparator();\n+\nprivate final HashMap<Long, List<MemoTableEntry>> _bestPlans =\nnew HashMap<Long, List<MemoTableEntry>>();\nprivate final HashSet<VisitMark> _visited = new HashSet<VisitMark>();\n@@ -84,6 +87,49 @@ public abstract class PlanSelection\n_visited.add(new VisitMark(hopID, type));\n}\n+ protected void rSelectPlansFuseAll(CPlanMemoTable memo, Hop current, TemplateType currentType, HashSet<Long> partition)\n+ {\n+ if( isVisited(current.getHopID(), currentType)\n+ || (partition!=null && !partition.contains(current.getHopID())) )\n+ return;\n+\n+ //step 1: prune subsumed plans of same type\n+ if( memo.contains(current.getHopID()) ) {\n+ HashSet<MemoTableEntry> rmSet = new HashSet<MemoTableEntry>();\n+ List<MemoTableEntry> hopP = memo.get(current.getHopID());\n+ for( MemoTableEntry e1 : hopP )\n+ for( MemoTableEntry e2 : hopP )\n+ if( e1 != e2 && e1.subsumes(e2) )\n+ rmSet.add(e2);\n+ memo.remove(current, rmSet);\n+ }\n+\n+ //step 2: select plan for current path\n+ MemoTableEntry best = null;\n+ if( memo.contains(current.getHopID()) ) {\n+ if( currentType == null ) {\n+ best = memo.get(current.getHopID()).stream()\n+ .filter(p -> isValid(p, current))\n+ .min(BASE_COMPARE).orElse(null);\n+ }\n+ else {\n+ _typedCompare.setType(currentType);\n+ best = memo.get(current.getHopID()).stream()\n+ .filter(p -> p.type==currentType || p.type==TemplateType.CELL)\n+ .min(_typedCompare).orElse(null);\n+ }\n+ addBestPlan(current.getHopID(), best);\n+ }\n+\n+ //step 3: recursively process children\n+ for( int i=0; i< current.getInput().size(); i++ ) {\n+ TemplateType pref = (best!=null && best.isPlanRef(i))? best.type : null;\n+ rSelectPlansFuseAll(memo, current.getInput().get(i), pref, partition);\n+ }\n+\n+ setVisited(current.getHopID(), currentType);\n+ }\n+\n/**\n* Basic plan comparator to compare memo table entries with regard to\n* a pre-defined template preference order and the number of references.\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/opt/PlanSelectionFuseAll.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/opt/PlanSelectionFuseAll.java", "diff": "package org.apache.sysml.hops.codegen.opt;\nimport java.util.ArrayList;\n-import java.util.Comparator;\nimport java.util.Map.Entry;\n-import java.util.HashSet;\nimport java.util.List;\nimport org.apache.sysml.hops.Hop;\nimport org.apache.sysml.hops.codegen.template.CPlanMemoTable;\nimport org.apache.sysml.hops.codegen.template.CPlanMemoTable.MemoTableEntry;\n-import org.apache.sysml.hops.codegen.template.TemplateBase.TemplateType;\n/**\n* This plan selection heuristic aims for maximal fusion, which\n@@ -43,52 +40,10 @@ public class PlanSelectionFuseAll extends PlanSelection\npublic void selectPlans(CPlanMemoTable memo, ArrayList<Hop> roots) {\n//pruning and collection pass\nfor( Hop hop : roots )\n- rSelectPlans(memo, hop, null);\n+ rSelectPlansFuseAll(memo, hop, null, null);\n//take all distinct best plans\nfor( Entry<Long, List<MemoTableEntry>> e : getBestPlans().entrySet() )\nmemo.setDistinct(e.getKey(), e.getValue());\n}\n-\n- private void rSelectPlans(CPlanMemoTable memo, Hop current, TemplateType currentType)\n- {\n- if( isVisited(current.getHopID(), currentType) )\n- return;\n-\n- //step 1: prune subsumed plans of same type\n- if( memo.contains(current.getHopID()) ) {\n- HashSet<MemoTableEntry> rmSet = new HashSet<MemoTableEntry>();\n- List<MemoTableEntry> hopP = memo.get(current.getHopID());\n- for( MemoTableEntry e1 : hopP )\n- for( MemoTableEntry e2 : hopP )\n- if( e1 != e2 && e1.subsumes(e2) )\n- rmSet.add(e2);\n- memo.remove(current, rmSet);\n- }\n-\n- //step 2: select plan for current path\n- MemoTableEntry best = null;\n- if( memo.contains(current.getHopID()) ) {\n- if( currentType == null ) {\n- best = memo.get(current.getHopID()).stream()\n- .filter(p -> isValid(p, current))\n- .min(new BasicPlanComparator()).orElse(null);\n- }\n- else {\n- best = memo.get(current.getHopID()).stream()\n- .filter(p -> p.type==currentType || p.type==TemplateType.CELL)\n- .min(Comparator.comparing(p -> 7-((p.type==currentType)?4:0)-p.countPlanRefs()))\n- .orElse(null);\n- }\n- addBestPlan(current.getHopID(), best);\n- }\n-\n- //step 3: recursively process children\n- for( int i=0; i< current.getInput().size(); i++ ) {\n- TemplateType pref = (best!=null && best.isPlanRef(i))? best.type : null;\n- rSelectPlans(memo, current.getInput().get(i), pref);\n- }\n-\n- setVisited(current.getHopID(), currentType);\n- }\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/opt/PlanSelectionFuseCostBased.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/opt/PlanSelectionFuseCostBased.java", "diff": "@@ -510,49 +510,6 @@ public class PlanSelectionFuseCostBased extends PlanSelection\nvisited.add(current.getHopID());\n}\n- private void rSelectPlansFuseAll(CPlanMemoTable memo, Hop current, TemplateType currentType, HashSet<Long> partition)\n- {\n- if( isVisited(current.getHopID(), currentType)\n- || !partition.contains(current.getHopID()) )\n- return;\n-\n- //step 1: prune subsumed plans of same type\n- if( memo.contains(current.getHopID()) ) {\n- HashSet<MemoTableEntry> rmSet = new HashSet<MemoTableEntry>();\n- List<MemoTableEntry> hopP = memo.get(current.getHopID());\n- for( MemoTableEntry e1 : hopP )\n- for( MemoTableEntry e2 : hopP )\n- if( e1 != e2 && e1.subsumes(e2) )\n- rmSet.add(e2);\n- memo.remove(current, rmSet);\n- }\n-\n- //step 2: select plan for current path\n- MemoTableEntry best = null;\n- if( memo.contains(current.getHopID()) ) {\n- if( currentType == null ) {\n- best = memo.get(current.getHopID()).stream()\n- .filter(p -> isValid(p, current))\n- .min(new BasicPlanComparator()).orElse(null);\n- }\n- else {\n- best = memo.get(current.getHopID()).stream()\n- .filter(p -> p.type==currentType || p.type==TemplateType.CELL)\n- .min(Comparator.comparing(p -> 7-((p.type==currentType)?4:0)-p.countPlanRefs()))\n- .orElse(null);\n- }\n- addBestPlan(current.getHopID(), best);\n- }\n-\n- //step 3: recursively process children\n- for( int i=0; i< current.getInput().size(); i++ ) {\n- TemplateType pref = (best!=null && best.isPlanRef(i))? best.type : null;\n- rSelectPlansFuseAll(memo, current.getInput().get(i), pref, partition);\n- }\n-\n- setVisited(current.getHopID(), currentType);\n- }\n-\nprivate static boolean[] createAssignment(int len, int pos) {\nboolean[] ret = new boolean[len];\nint tmp = pos;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/opt/PlanSelectionFuseCostBasedV2.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/opt/PlanSelectionFuseCostBasedV2.java", "diff": "@@ -98,8 +98,6 @@ public class PlanSelectionFuseCostBasedV2 extends PlanSelection\nprivate static final IDSequence COST_ID = new IDSequence();\nprivate static final TemplateRow ROW_TPL = new TemplateRow();\n- private static final BasicPlanComparator BASE_COMPARE = new BasicPlanComparator();\n- private final TypedPlanComparator _typedCompare = new TypedPlanComparator();\n@Override\npublic void selectPlans(CPlanMemoTable memo, ArrayList<Hop> roots)\n@@ -729,49 +727,6 @@ public class PlanSelectionFuseCostBasedV2 extends PlanSelection\nvisited.add(current.getHopID());\n}\n- private void rSelectPlansFuseAll(CPlanMemoTable memo, Hop current, TemplateType currentType, HashSet<Long> partition)\n- {\n- if( isVisited(current.getHopID(), currentType)\n- || !partition.contains(current.getHopID()) )\n- return;\n-\n- //step 1: prune subsumed plans of same type\n- if( memo.contains(current.getHopID()) ) {\n- HashSet<MemoTableEntry> rmSet = new HashSet<MemoTableEntry>();\n- List<MemoTableEntry> hopP = memo.get(current.getHopID());\n- for( MemoTableEntry e1 : hopP )\n- for( MemoTableEntry e2 : hopP )\n- if( e1 != e2 && e1.subsumes(e2) )\n- rmSet.add(e2);\n- memo.remove(current, rmSet);\n- }\n-\n- //step 2: select plan for current path\n- MemoTableEntry best = null;\n- if( memo.contains(current.getHopID()) ) {\n- if( currentType == null ) {\n- best = memo.get(current.getHopID()).stream()\n- .filter(p -> isValid(p, current))\n- .min(BASE_COMPARE).orElse(null);\n- }\n- else {\n- _typedCompare.setType(currentType);\n- best = memo.get(current.getHopID()).stream()\n- .filter(p -> p.type==currentType || p.type==TemplateType.CELL)\n- .min(_typedCompare).orElse(null);\n- }\n- addBestPlan(current.getHopID(), best);\n- }\n-\n- //step 3: recursively process children\n- for( int i=0; i< current.getInput().size(); i++ ) {\n- TemplateType pref = (best!=null && best.isPlanRef(i))? best.type : null;\n- rSelectPlansFuseAll(memo, current.getInput().get(i), pref, partition);\n- }\n-\n- setVisited(current.getHopID(), currentType);\n- }\n-\n/////////////////////////////////////////////////////////\n// Cost model fused operators w/ materialization points\n//////////\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateUtils.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateUtils.java", "diff": "@@ -204,6 +204,8 @@ public class TemplateUtils\nreturn RowType.COL_AGG_B1_T;\nelse if( B1 != null && output.getDim1()==B1.getDim2() && output.getDim2()==X.getDim2())\nreturn RowType.COL_AGG_B1;\n+ else if( B1 != null && output.getDim1()==1 && B1.getDim2() == output.getDim2() )\n+ return RowType.COL_AGG_B1R;\nelse if( X.getDim1() == output.getDim1() && X.getDim2() != output.getDim2() )\nreturn RowType.NO_AGG_CONST;\nelse\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofRowwise.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofRowwise.java", "diff": "@@ -55,14 +55,17 @@ public abstract class SpoofRowwise extends SpoofOperator\nCOL_AGG, //col aggregation (e.g., colSums() or t(y) %*% X)\nCOL_AGG_T, //transposed col aggregation (e.g., t(X) %*% y)\nCOL_AGG_B1, //col aggregation w/ matrix mult B1\n- COL_AGG_B1_T; //transposed col aggregation w/ matrix mult B1\n+ COL_AGG_B1_T, //transposed col aggregation w/ matrix mult B1\n+ COL_AGG_B1R; //col aggregation w/ matrix mult B1 to row vector\npublic boolean isColumnAgg() {\n- return (this == COL_AGG || this == COL_AGG_T)\n- || (this == COL_AGG_B1) || (this == COL_AGG_B1_T);\n+ return this == COL_AGG || this == COL_AGG_T\n+ || this == COL_AGG_B1 || this == COL_AGG_B1_T\n+ || this == COL_AGG_B1R;\n}\npublic boolean isRowTypeB1() {\n- return (this == NO_AGG_B1) || (this == COL_AGG_B1) || (this == COL_AGG_B1_T);\n+ return this == NO_AGG_B1 || this == COL_AGG_B1\n+ || this == COL_AGG_B1_T || this == COL_AGG_B1R;\n}\npublic boolean isRowTypeB1ColumnAgg() {\nreturn (this == COL_AGG_B1) || (this == COL_AGG_B1_T);\n@@ -268,7 +271,7 @@ public abstract class SpoofRowwise extends SpoofOperator\ncase COL_AGG_T: out.reset(n, 1, false); break;\ncase COL_AGG_B1: out.reset(n2, n, false); break;\ncase COL_AGG_B1_T: out.reset(n, n2, false); break;\n-\n+ case COL_AGG_B1R: out.reset(1, n2, false); break;\n}\nout.allocateDenseBlock();\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/IndexingOp.java", "new_path": "src/main/java/org/apache/sysml/hops/IndexingOp.java", "diff": "@@ -237,7 +237,8 @@ public class IndexingOp extends Hop\n@Override\nprotected double computeOutputMemEstimate( long dim1, long dim2, long nnz )\n{\n- double sparsity = OptimizerUtils.getSparsity(dim1, dim2, nnz);\n+ // only dense right indexing supported on GPU\n+ double sparsity = isGPUEnabled() ? 1.0 : OptimizerUtils.getSparsity(dim1, dim2, nnz);\nreturn OptimizerUtils.estimateSizeExactSparsity(dim1, dim2, sparsity);\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/perftest/python/run_perftest.py", "new_path": "scripts/perftest/python/run_perftest.py", "diff": "@@ -283,7 +283,7 @@ if __name__ == '__main__':\ndefault_mat_shape = ['10k_100']\n# Default temp directory, contains everything generated in perftest\n- default_config_dir = join(systemml_home, 'scripts', 'perftest', 'temp')\n+ default_config_dir = join(systemml_home, 'temp_perftest')\n# Initialize time\nstart_time = time.time()\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "<goal>run</goal>\n</goals>\n<configuration>\n- <protocVersion>2.6.1</protocVersion> <!-- 2.4.1, 2.5.0, 2.6.1, 3.4.0 -->\n+ <protocVersion>2.5.0</protocVersion> <!-- 2.4.1, 2.5.0, 2.6.1, 3.4.0 -->\n<inputDirectories>\n<include>src/main/proto/caffe</include>\n</inputDirectories>\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/compress/CompressedLinregCG.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/compress/CompressedLinregCG.java", "diff": "@@ -55,6 +55,7 @@ public class CompressedLinregCG extends AutomatedTestBase\nprivate final static int intercept = 0;\nprivate final static double epsilon = 0.000000001;\nprivate final static double maxiter = 10;\n+ private final static double regular = 0.001;\n@Override\npublic void setUp() {\n@@ -103,20 +104,17 @@ public class CompressedLinregCG extends AutomatedTestBase\nTestConfiguration config = getTestConfiguration(TEST_NAME);\n/* This is for running the junit test the new way, i.e., construct the arguments directly */\n- String HOME = SCRIPT_DIR + TEST_DIR;\n- fullDMLScriptName = HOME + TEST_NAME + \".dml\";\n- programArgs = new String[]{ \"-explain\",\"-stats\",\n- \"-args\", HOME + INPUT_DIR + \"X\",\n- HOME + INPUT_DIR + \"y\",\n- String.valueOf(intercept),\n- String.valueOf(epsilon),\n- String.valueOf(maxiter),\n- HOME + OUTPUT_DIR + \"w\"};\n- fullRScriptName = HOME + TEST_NAME + \".R\";\n+ String HOME = SCRIPT_DIR + \"functions/codegen/\";\n+ fullDMLScriptName = \"scripts/algorithms/LinearRegCG.dml\";\n+ programArgs = new String[]{ \"-explain\", \"-stats\", \"-nvargs\", \"X=\"+input(\"X\"), \"Y=\"+input(\"y\"),\n+ \"icpt=\"+String.valueOf(intercept), \"tol=\"+String.valueOf(epsilon),\n+ \"maxi=\"+String.valueOf(maxiter), \"reg=\"+String.valueOf(regular), \"B=\"+output(\"w\")};\n+\n+ fullRScriptName = HOME + \"Algorithm_LinregCG.R\";\nrCmd = \"Rscript\" + \" \" + fullRScriptName + \" \" +\nHOME + INPUT_DIR + \" \" +\nString.valueOf(intercept) + \" \" + String.valueOf(epsilon) + \" \" +\n- String.valueOf(maxiter) + \" \" + HOME + EXPECTED_DIR;\n+ String.valueOf(maxiter) + \" \" + String.valueOf(regular) + HOME + EXPECTED_DIR;\nloadTestConfiguration(config);\n" }, { "change_type": "DELETE", "old_path": "src/test/scripts/functions/compress/LinregCG.R", "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-\n-args <- commandArgs(TRUE)\n-options(digits=22)\n-library(\"Matrix\")\n-\n-X = readMM(paste(args[1], \"X.mtx\", sep=\"\"))\n-y = readMM(paste(args[1], \"y.mtx\", sep=\"\"))\n-\n-intercept = as.integer(args[2]);\n-eps = as.double(args[3]);\n-maxiter = as.double(args[4]);\n-\n-if( intercept == 1 ){\n- ones = matrix(1, nrow(X), 1);\n- X = cbind(X, ones);\n-}\n-\n-r = -(t(X) %*% y);\n-p = -r;\n-norm_r2 = sum(r * r);\n-w = matrix(0, ncol(X), 1);\n-\n-i = 0;\n-while(i < maxiter) {\n- q = ((t(X) %*% (X %*% p)) + eps * p);\n- alpha = norm_r2 / ((t(p) %*% q)[1:1]);\n- w = w + alpha * p;\n- old_norm_r2 = norm_r2;\n- r = r + alpha * q;\n- norm_r2 = sum(r * r);\n- beta = norm_r2 / old_norm_r2;\n- p = -r + beta * p;\n- i = i + 1;\n-}\n-\n-writeMM(as(w,\"CsparseMatrix\"), paste(args[5], \"w\", sep=\"\"))\n" }, { "change_type": "DELETE", "old_path": "src/test/scripts/functions/compress/LinregCG.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-\n-X = read($1);\n-y = read($2);\n-intercept = $3;\n-eps = $4;\n-maxiter = $5;\n-\n-if( intercept == 1 ){\n- ones = matrix(1, nrow(X), 1);\n- X = cbind(X, ones);\n-}\n-\n-r = -(t(X) %*% y);\n-p = -r;\n-norm_r2 = sum(r * r);\n-w = matrix(0, rows = ncol(X), cols = 1);\n-\n-i = 0;\n-while(i < maxiter) {\n- q = ((t(X) %*% (X %*% p)) + eps * p);\n- alpha = norm_r2 / castAsScalar(t(p) %*% q);\n- w = w + alpha * p;\n- old_norm_r2 = norm_r2;\n- r = r + alpha * q;\n- norm_r2 = sum(r * r);\n- beta = norm_r2 / old_norm_r2;\n- p = -r + beta * p;\n- i = i + 1;\n-}\n-\n-write(w, $6);\n-\n-\n-\n-\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/parser/StatementBlock.java", "new_path": "src/main/java/org/apache/sysml/parser/StatementBlock.java", "diff": "@@ -545,24 +545,19 @@ public class StatementBlock extends LiveVariableAnalysis implements ParseInfo\nthrows LanguageException, ParseException, IOException\n{\n_constVarsIn.putAll(constVars);\n- HashMap<String, ConstIdentifier> currConstVars = new HashMap<String,ConstIdentifier>();\n- currConstVars.putAll(constVars);\n-\n_statements = rewriteFunctionCallStatements(dmlProg, _statements);\n_dmlProg = dmlProg;\n+ HashMap<String, ConstIdentifier> currConstVars = new HashMap<String,ConstIdentifier>(constVars);\nfor (Statement current : _statements) {\n-\nif (current instanceof OutputStatement) {\nOutputStatement os = (OutputStatement)current;\n-\n// validate variable being written by output statement exists\nDataIdentifier target = (DataIdentifier)os.getIdentifier();\nif (ids.getVariable(target.getName()) == null) {\n//undefined variables are always treated unconditionally as error in order to prevent common script-level bugs\nraiseValidateError(\"Undefined Variable (\" + target.getName() + \") used in statement\", false, LanguageErrorCodes.INVALID_PARAMETERS);\n}\n-\nif ( ids.getVariable(target.getName()).getDataType() == DataType.SCALAR) {\nboolean paramsOkay = true;\nfor (String key : os.getSource().getVarParams().keySet()){\n@@ -577,20 +572,59 @@ public class StatementBlock extends LiveVariableAnalysis implements ParseInfo\nExpression source = os.getSource();\nsource.setOutput(target);\nsource.validateExpression(ids.getVariables(), currConstVars, conditional);\n-\nsetStatementFormatType(os, conditional);\ntarget.setDimensionValueProperties(ids.getVariable(target.getName()));\n}\n-\nelse if (current instanceof AssignmentStatement){\n+ validateAssignmentStatement(current, dmlProg, ids, currConstVars, conditional);\n+ }\n+ else if (current instanceof MultiAssignmentStatement){\n+ validateMultiAssignmentStatement(current, dmlProg, ids, currConstVars, conditional);\n+ }\n+ else if(current instanceof ForStatement || current instanceof IfStatement || current instanceof WhileStatement ){\n+ raiseValidateError(\"control statement (WhileStatement, IfStatement, ForStatement) should not be in generic statement block. Likely a parsing error\", conditional);\n+ }\n+ else if (current instanceof PrintStatement) {\n+ PrintStatement pstmt = (PrintStatement) current;\n+ List<Expression> expressions = pstmt.getExpressions();\n+ for (Expression expression : expressions) {\n+ expression.validateExpression(ids.getVariables(), currConstVars, conditional);\n+ if (expression.getOutput().getDataType() != Expression.DataType.SCALAR) {\n+ if (expression.getOutput().getDataType() == Expression.DataType.MATRIX) {\n+ pstmt.raiseValidateError(\"Print statements can only print scalars. To print a matrix, please wrap it in a toString() function.\", conditional);\n+ } else {\n+ pstmt.raiseValidateError(\"Print statements can only print scalars.\", conditional);\n+ }\n+ }\n+ }\n+ }\n+ // no work to perform for PathStatement or ImportStatement\n+ else if (current instanceof PathStatement){}\n+ else if (current instanceof ImportStatement){}\n+ else {\n+ raiseValidateError(\"cannot process statement of type \" + current.getClass().getSimpleName(), conditional);\n+ }\n+ }\n+ _constVarsOut.putAll(currConstVars);\n+ return ids;\n+ }\n+\n+ private void validateAssignmentStatement(Statement current, DMLProgram dmlProg,\n+ VariableSet ids, HashMap<String, ConstIdentifier> currConstVars, boolean conditional)\n+ throws LanguageException, IOException, ParseException\n+ {\nAssignmentStatement as = (AssignmentStatement)current;\nDataIdentifier target = as.getTarget();\nExpression source = as.getSource();\nif (source instanceof FunctionCallIdentifier) {\n- ((FunctionCallIdentifier) source).validateExpression(dmlProg, ids.getVariables(),currConstVars, conditional);\n+ ((FunctionCallIdentifier) source).validateExpression(\n+ dmlProg, ids.getVariables(),currConstVars, conditional);\n}\n- else {\n+ else { //all builtin functions and expressions\n+ if( target == null )\n+ raiseValidateError(\"Missing variable assignment.\", false);\n+\nif( MLContextProxy.isActive() )\nMLContextProxy.setAppropriateVarsForRead(source, target._name);\n@@ -636,22 +670,19 @@ public class StatementBlock extends LiveVariableAnalysis implements ParseInfo\n// handle case when nrow / ncol called on variable with size unknown (dims == -1)\n// --> const prop NOT possible\n- if (intid.getValue() != -1){\n+ if (intid.getValue() != -1)\ncurrConstVars.put(target.getName(), intid);\n}\n}\n- }\nif (target == null) {\n// function has no return value\n}\n// CASE: target NOT indexed identifier\nelse if (!(target instanceof IndexedIdentifier)){\ntarget.setProperties(source.getOutput());\n- if (source.getOutput() instanceof IndexedIdentifier){\n+ if (source.getOutput() instanceof IndexedIdentifier)\ntarget.setDimensions(source.getOutput().getDim1(), source.getOutput().getDim2());\n}\n-\n- }\n// CASE: target is indexed identifier\nelse\n{\n@@ -672,12 +703,6 @@ public class StatementBlock extends LiveVariableAnalysis implements ParseInfo\nif ( ((IndexedIdentifier)target).getColUpperBound() != null )\n((IndexedIdentifier)target).getColUpperBound().validateExpression(ids.getVariables(), currConstVars, conditional);\n- // validate that LHS indexed identifier is being assigned a matrix value\n-// if (source.getOutput().getDataType() != Expression.DataType.MATRIX){\n-// LOG.error(target.printErrorLocation() + \"Indexed expression \" + target.toString() + \" can only be assigned matrix value\");\n-// throw new LanguageException(target.printErrorLocation() + \"Indexed expression \" + target.toString() + \" can only be assigned matrix value\");\n-// }\n-\n// validate that size of LHS index ranges is being assigned:\n// (a) a matrix value of same size as LHS\n// (b) singleton value (semantics: initialize enitre submatrix with this value)\n@@ -694,33 +719,27 @@ public class StatementBlock extends LiveVariableAnalysis implements ParseInfo\n+ targetSize._row + \" rows and \" + targetSize._col + \" cols. Attempted to assign matrix with dimensions \"\n+ source.getOutput().getDim1() + \" rows and \" + source.getOutput().getDim2() + \" cols \", conditional);\n}\n-\n((IndexedIdentifier)target).setDimensions(targetSize._row, targetSize._col);\n}\n- if (target != null) {\n+ if (target != null)\nids.addVariable(target.getName(), target);\n}\n- }\n-\n- else if (current instanceof MultiAssignmentStatement){\n+ private void validateMultiAssignmentStatement(Statement current, DMLProgram dmlProg,\n+ VariableSet ids, HashMap<String, ConstIdentifier> currConstVars, boolean conditional)\n+ throws LanguageException, IOException\n+ {\nMultiAssignmentStatement mas = (MultiAssignmentStatement) current;\nArrayList<DataIdentifier> targetList = mas.getTargetList();\n-\n- // perform validation of source expression\nExpression source = mas.getSource();\n- /*\n- * MultiAssignmentStatments currently supports only External,\n- * User-defined, and Multi-return Builtin function expressions\n- */\n+\n+ //MultiAssignmentStatments currently supports only External,\n+ //User-defined, and Multi-return Builtin function expressions\nif (!(source instanceof DataIdentifier)\n|| (source instanceof DataIdentifier && !((DataIdentifier)source).multipleReturns()) ) {\n- //if (!(source instanceof FunctionCallIdentifier) ) {\n- //|| !(source instanceof BuiltinFunctionExpression && ((BuiltinFunctionExpression)source).isMultiReturnBuiltinFunction()) ){\nsource.raiseValidateError(\"can only use user-defined functions with multi-assignment statement\", conditional);\n}\n-\nif ( source instanceof FunctionCallIdentifier) {\nFunctionCallIdentifier fci = (FunctionCallIdentifier)source;\nfci.validateExpression(dmlProg, ids.getVariables(), currConstVars, conditional);\n@@ -732,16 +751,16 @@ public class StatementBlock extends LiveVariableAnalysis implements ParseInfo\nelse\nthrow new LanguageException(\"Unexpected error.\");\n-\nif ( source instanceof FunctionCallIdentifier ) {\nfor (int j =0; j< targetList.size(); j++) {\n-\nDataIdentifier target = targetList.get(j);\n// set target properties (based on type info in function call statement return params)\nFunctionCallIdentifier fci = (FunctionCallIdentifier)source;\n- FunctionStatement fstmt = (FunctionStatement)_dmlProg.getFunctionStatementBlock(fci.getNamespace(), fci.getName()).getStatement(0);\n+ FunctionStatement fstmt = (FunctionStatement)_dmlProg\n+ .getFunctionStatementBlock(fci.getNamespace(), fci.getName()).getStatement(0);\nif (fstmt == null){\n- fci.raiseValidateError(\" function \" + fci.getName() + \" is undefined in namespace \" + fci.getNamespace(), conditional);\n+ fci.raiseValidateError(\" function \" + fci.getName()\n+ + \" is undefined in namespace \" + fci.getNamespace(), conditional);\n}\nif (!(target instanceof IndexedIdentifier)){\ntarget.setProperties(fstmt.getOutputParams().get(j));\n@@ -749,7 +768,8 @@ public class StatementBlock extends LiveVariableAnalysis implements ParseInfo\nelse{\nDataIdentifier targetAsSeen = ids.getVariable(target.getName());\nif (targetAsSeen == null){\n- raiseValidateError(target.printErrorLocation() + \"cannot assign value to indexed identifier \" + target.toString() + \" without first initializing \" + target.getName(), conditional);\n+ raiseValidateError(target.printErrorLocation() + \"cannot assign value to indexed identifier \"\n+ + target.toString() + \" without first initializing \" + target.getName(), conditional);\n}\ntarget.setProperties(targetAsSeen);\n}\n@@ -764,40 +784,6 @@ public class StatementBlock extends LiveVariableAnalysis implements ParseInfo\n}\n}\n- else if(current instanceof ForStatement || current instanceof IfStatement || current instanceof WhileStatement ){\n- raiseValidateError(\"control statement (WhileStatement, IfStatement, ForStatement) should not be in generic statement block. Likely a parsing error\", conditional);\n- }\n-\n- else if (current instanceof PrintStatement) {\n- PrintStatement pstmt = (PrintStatement) current;\n- List<Expression> expressions = pstmt.getExpressions();\n- for (Expression expression : expressions) {\n- expression.validateExpression(ids.getVariables(), currConstVars, conditional);\n- if (expression.getOutput().getDataType() != Expression.DataType.SCALAR) {\n- if (expression.getOutput().getDataType() == Expression.DataType.MATRIX) {\n- pstmt.raiseValidateError(\"Print statements can only print scalars. To print a matrix, please wrap it in a toString() function.\", conditional);\n- } else {\n- pstmt.raiseValidateError(\"Print statements can only print scalars.\", conditional);\n- }\n- }\n- }\n- }\n-\n- // no work to perform for PathStatement or ImportStatement\n- else if (current instanceof PathStatement){}\n- else if (current instanceof ImportStatement){}\n-\n-\n- else {\n- raiseValidateError(\"cannot process statement of type \" + current.getClass().getSimpleName(), conditional);\n- }\n-\n- } // end for (Statement current : _statements){\n- _constVarsOut.putAll(currConstVars);\n- return ids;\n-\n- }\n-\npublic void setStatementFormatType(OutputStatement s, boolean conditionalValidate)\nthrows LanguageException, ParseException\n{\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/java/org/apache/sysml/test/integration/functions/misc/InvalidBuiltinFunctionCallTest.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.sysml.test.integration.functions.misc;\n+\n+import org.junit.Test;\n+\n+import org.apache.sysml.api.DMLException;\n+import org.apache.sysml.test.integration.AutomatedTestBase;\n+import org.apache.sysml.test.integration.TestConfiguration;\n+import org.apache.sysml.test.utils.TestUtils;\n+\n+public class InvalidBuiltinFunctionCallTest extends AutomatedTestBase\n+{\n+ private final static String TEST_DIR = \"functions/misc/\";\n+ private final static String TEST_NAME1 = \"InvalidBuiltinFunctionCallTest1\";\n+ private final static String TEST_CLASS_DIR = TEST_DIR + InvalidBuiltinFunctionCallTest.class.getSimpleName() + \"/\";\n+\n+ @Override\n+ public void setUp() {\n+ TestUtils.clearAssertionInformation();\n+ addTestConfiguration(TEST_NAME1, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME1, new String[] {}));\n+ }\n+\n+ @Test\n+ public void testInvalidBuiltinFunctionCall1() {\n+ runTest( TEST_NAME1, true );\n+ }\n+\n+ private void runTest( String testName, boolean expected )\n+ {\n+ TestConfiguration config = getTestConfiguration(testName);\n+ loadTestConfiguration(config);\n+\n+ String HOME = SCRIPT_DIR + TEST_DIR;\n+ fullDMLScriptName = HOME + testName + \".dml\";\n+ programArgs = new String[]{};\n+\n+ //run tests\n+ runTest(true, expected, DMLException.class, -1);\n+ }\n+}\n" }, { "change_type": "ADD", "old_path": null, "new_path": "src/test/scripts/functions/misc/InvalidBuiltinFunctionCallTest1.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 = rand(rows=1006, cols=784, sparsity=0.001);\n+removeEmpty(target=X, margin=\"rows\");\n+print(\"nrow(X): \" + nrow(X));\n" }, { "change_type": "MODIFY", "old_path": "src/test_suites/java/org/apache/sysml/test/integration/functions/misc/ZPackageSuite.java", "new_path": "src/test_suites/java/org/apache/sysml/test/integration/functions/misc/ZPackageSuite.java", "diff": "@@ -32,6 +32,7 @@ import org.junit.runners.Suite;\nFunctionInliningTest.class,\nFunctionNamespaceTest.class,\nIfTest.class,\n+ InvalidBuiltinFunctionCallTest.class,\nInvalidFunctionAssignmentTest.class,\nInvalidFunctionSignatureTest.class,\nIPAConstantFoldingScalarVariablePropagationTest.class,\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeBinary.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/cplan/CNodeBinary.java", "diff": "@@ -269,10 +269,9 @@ public class CNodeBinary extends CNode\nsb.append(_inputs.get(1).codegen(sparse));\n//generate binary operation (use sparse template, if data input)\n- boolean lsparse = sparse && (_inputs.get(0) instanceof CNodeData\n- && (_inputs.get(0).getVarname().startsWith(\"a\")\n- || _inputs.get(1).getVarname().startsWith(\"a\"))\n- && !_inputs.get(0).isLiteral());\n+ boolean lsparse = sparse\n+ && ((_inputs.get(0) instanceof CNodeData && _inputs.get(0).getVarname().startsWith(\"a\"))\n+ ||(_inputs.get(1) instanceof CNodeData && _inputs.get(1).getVarname().startsWith(\"a\")));\nboolean scalarInput = _inputs.get(0).getDataType().isScalar();\nboolean scalarVector = (_inputs.get(0).getDataType().isScalar()\n&& _inputs.get(1).getDataType().isMatrix());\n" } ]

[ { "change_type": "MODIFY", "old_path": "bin/systemml-spark-submit.py", "new_path": "bin/systemml-spark-submit.py", "diff": "@@ -41,8 +41,8 @@ def default_jars(systemml_home):\nreturn target_jars, systemml_jar\n-def spark_submit_entry(master, driver_memory, num_executors, executor_memory,\n- executor_cores, conf,\n+def spark_submit_entry(master, deploy_mode, driver_memory, num_executors,\n+ executor_memory, executor_cores, conf,\nnvargs, args, config, explain, debug, stats, gpu, f):\n\"\"\"\nThis function is responsible for the execution of arguments via\n@@ -100,7 +100,7 @@ def spark_submit_entry(master, driver_memory, num_executors, executor_memory,\n# stats, explain, target_jars\ncmd_spark = [spark_path, '--class', 'org.apache.sysml.api.DMLScript',\n- '--master', master,\n+ '--master', master, '--deploy-mode', deploy_mode,\n'--driver-memory', driver_memory,\n'--conf', default_conf,\n'--jars', cuda_jars, systemml_jars]\n@@ -129,7 +129,8 @@ if __name__ == '__main__':\ncparser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter,\ndescription='System-ML Spark Submit Script')\n# SPARK-SUBMIT Options\n- cparser.add_argument('--master', default='local[*]', help='local, yarn-client, yarn-cluster', metavar='')\n+ cparser.add_argument('--master', default='local[*]', help='local, yarn', metavar='')\n+ cparser.add_argument('--deploy-mode', help='client, cluster', default='client', metavar='')\ncparser.add_argument('--driver-memory', default='8G', help='Memory for driver (e.g. 512M, 1G)', metavar='')\ncparser.add_argument('--num-executors', nargs=1, help='Number of executors to launch', metavar='')\ncparser.add_argument('--executor-memory', nargs=1, help='Memory per executor', metavar='')\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/run_perftest.py", "new_path": "scripts/perftest/python/run_perftest.py", "diff": "@@ -355,7 +355,8 @@ if __name__ == '__main__':\n'set <force> option to skip conservative memory estimates '\n'and use GPU wherever possible', nargs='?', const='no_option')\n# Spark Configuration Option\n- cparser.add_argument('--master', help='local, yarn-client, yarn-cluster', metavar='')\n+ cparser.add_argument('--master', help='local, yarn', metavar='')\n+ cparser.add_argument('--deploy-mode', help='client, cluster', metavar='')\ncparser.add_argument('--driver-memory', help='Memory for driver (e.g. 512M)', metavar='')\ncparser.add_argument('--num-executors', help='Number of executors to launch', metavar='')\ncparser.add_argument('--executor-memory', help='Memory per executor', metavar='')\n" }, { "change_type": "MODIFY", "old_path": "scripts/perftest/python/utils_misc.py", "new_path": "scripts/perftest/python/utils_misc.py", "diff": "@@ -84,6 +84,9 @@ def split_config_args(args):\nif args['master'] is not None:\nbackend_args_dict['--master'] = args['master']\n+ if args['deploy_mode'] is not None:\n+ backend_args_dict['--deploy-mode'] = args['master']\n+\nif args['num_executors'] is not None:\nbackend_args_dict['--num-executors'] = args['num_executors']\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/rewrite/RewriteElementwiseMultChainOptimization.java", "new_path": "src/main/java/org/apache/sysml/hops/rewrite/RewriteElementwiseMultChainOptimization.java", "diff": "@@ -310,7 +310,7 @@ public class RewriteElementwiseMultChainOptimization extends HopRewriteRule {\nfinal ArrayList<Hop> parents = child.getParent();\nif (parents.size() > 1)\nfor (final Hop parent : parents)\n- if (parent instanceof BinaryOp && !emults.contains(parent))\n+ if (!(parent instanceof BinaryOp) || !emults.contains(parent))\nreturn false;\n// child does not have foreign parents\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/FrameBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/FrameBlock.java", "diff": "@@ -553,6 +553,20 @@ public class FrameBlock implements Writable, CacheBlock, Externalizable\nreturn new ObjectRowIterator(0, _numRows);\n}\n+ /**\n+ * Get a row iterator over the frame where all fields are encoded\n+ * as boxed objects according to the value types of the provided\n+ * target schema.\n+ *\n+ * @param schema target schema of objects\n+ * @return object array iterator\n+ */\n+ public Iterator<Object[]> getObjectRowIterator(ValueType[] schema) {\n+ ObjectRowIterator iter = new ObjectRowIterator(0, _numRows);\n+ iter.setSchema(schema);\n+ return iter;\n+ }\n+\n/**\n* Get a row iterator over the frame where all selected fields are\n* encoded as boxed objects according to their value types.\n@@ -992,7 +1006,7 @@ public class FrameBlock implements Writable, CacheBlock, Externalizable\nret._coldata = new Array[getNumColumns()];\nfor( int j=0; j<getNumColumns(); j++ )\nret._coldata[j] = _coldata[j].clone();\n- Iterator<Object[]> iter = that.getObjectRowIterator();\n+ Iterator<Object[]> iter = that.getObjectRowIterator(_schema);\nwhile( iter.hasNext() )\nret.appendRow(iter.next());\n}\n@@ -1221,6 +1235,8 @@ public class FrameBlock implements Writable, CacheBlock, Externalizable\n}\nprivate class ObjectRowIterator extends RowIterator<Object> {\n+ private ValueType[] _tgtSchema = null;\n+\npublic ObjectRowIterator(int rl, int ru) {\nsuper(rl, ru);\n}\n@@ -1229,6 +1245,10 @@ public class FrameBlock implements Writable, CacheBlock, Externalizable\nsuper(rl, ru, cols);\n}\n+ public void setSchema(ValueType[] schema) {\n+ _tgtSchema = schema;\n+ }\n+\n@Override\nprotected Object[] createRow(int size) {\nreturn new Object[size];\n@@ -1237,10 +1257,17 @@ public class FrameBlock implements Writable, CacheBlock, Externalizable\n@Override\npublic Object[] next( ) {\nfor( int j=0; j<_cols.length; j++ )\n- _curRow[j] = get(_curPos, _cols[j]-1);\n+ _curRow[j] = getValue(_curPos, _cols[j]-1);\n_curPos++;\nreturn _curRow;\n}\n+\n+ private Object getValue(int i, int j) {\n+ Object val = get(i, j);\n+ if( _tgtSchema != null )\n+ val = UtilFunctions.objectToObject(_tgtSchema[j], val);\n+ return val;\n+ }\n}\n///////\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixReorg.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixReorg.java", "diff": "@@ -859,8 +859,8 @@ public class LibMatrixReorg\nif( cl > 0 ) {\nfor( int i=bi; i<bimin; i++ )\nif( !a.isEmpty(i) ) {\n- int pos = a.posFIndexGTE(i, cl);\n- ix[i-bi] = (pos>=0) ? pos : a.size(i);\n+ int j = a.posFIndexGTE(i, cl);\n+ ix[i-bi] = (j>=0) ? j : a.size(i);\n}\n}\n@@ -868,19 +868,19 @@ public class LibMatrixReorg\nint bjmin = Math.min(bj+blocksizeJ, cu);\n//core block transpose operation\n- for( int i=bi, iix=0; i<bimin; i++, iix++ ) {\n+ for( int i=bi; i<bimin; i++ ) {\nif( a.isEmpty(i) ) continue;\nint apos = a.pos(i);\nint alen = a.size(i);\nint[] aix = a.indexes(i);\ndouble[] avals = a.values(i);\n- int j = ix[iix]; //last block boundary\n- for( ; j<alen && aix[j]<bjmin; j++ ) {\n- c.allocate(aix[apos+j], ennz2,n2);\n- c.append(aix[apos+j], i, avals[apos+j]);\n+ int j = ix[i-bi] + apos; //last block boundary\n+ for( ; j<apos+alen && aix[j]<bjmin; j++ ) {\n+ c.allocate(aix[j], ennz2,n2);\n+ c.append(aix[j], i, avals[j]);\n}\n- ix[iix] = j; //keep block boundary\n+ ix[i-bi] = j - apos; //keep block boundary\n}\n}\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/datagen/genRandData4ALS.dml", "new_path": "scripts/datagen/genRandData4ALS.dml", "diff": "#-------------------------------------------------------------\nXfile = $X; # input matrix X of size m x n\n-Wfile = $W; # original row factor of size m x r\n-Hfile = $H; # original col factor of size r x n\n+Ufile = $U; # original row factor of size m x r\n+Vfile = $V; # original col factor of size r x n\nm = $rows; # no. of rows of X\nn = $cols; # no. of cols of X\nr = $rank; # rank of factorization\n@@ -30,15 +30,18 @@ sigma = ifdef ($sigma, 0.01); # variance of Gaussian noise\nfmt = ifdef ($fmt, \"binary\"); # output format\n# generate original factors by sampling from a normal(0,1.0) distribution\n-W = rand(rows = m, cols = r, pdf = \"normal\", seed = 123);\n-H = rand(rows = r, cols = n, pdf = \"normal\", seed = 456);\n+U = rand(rows = m, cols = r, pdf = \"normal\", seed = 123);\n+V = rand(rows = n, cols = r, pdf = \"normal\", seed = 456);\nI = floor(rand(rows = nnz, cols = 1, min = 1, max = m + 0.999999999));\nJ = floor(rand(rows = nnz, cols = 1, min = 1, max = n + 0.999999999));\nX = rand(rows = nnz, cols = 1, pdf = \"normal\") * sqrt(sigma);\nN = table(I, J, X);\n-T = (N != 0);\n-X = T * (W %*% H) + T * N;\n+X = (N != 0) * (U %*% t(V)) + N;\nwrite(X, Xfile, format = fmt);\n-write(W, Wfile, format = fmt);\n-write(H, Hfile, format = fmt);\n+if( Ufile != \" \" )\n+ write(U, Ufile, format = fmt);\n+if( Vfile != \" \" ) {\n+ V = t(V);\n+ write(V, Vfile, format = fmt);\n+}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/ConvolutionOp.java", "new_path": "src/main/java/org/apache/sysml/hops/ConvolutionOp.java", "diff": "@@ -639,6 +639,9 @@ public class ConvolutionOp extends Hop implements MultiThreadedHop\nreturn;\n}\n+ // Reset the _cachedParams to avoid incorrect sizes\n+ _cachedParams = new ConvolutionParameters(-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, _maxNumThreads);\n+\nswitch(op)\n{\ncase MAX_POOLING:\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/ParameterizedBuiltinCPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/ParameterizedBuiltinCPInstruction.java", "diff": "@@ -26,6 +26,7 @@ import org.apache.sysml.lops.Lop;\nimport org.apache.sysml.parser.ParameterizedBuiltinFunctionExpression;\nimport org.apache.sysml.parser.Statement;\nimport org.apache.sysml.runtime.DMLRuntimeException;\n+import org.apache.sysml.runtime.controlprogram.caching.CacheBlock;\nimport org.apache.sysml.runtime.controlprogram.caching.CacheableData;\nimport org.apache.sysml.runtime.controlprogram.caching.FrameObject;\nimport org.apache.sysml.runtime.controlprogram.caching.MatrixObject;\n@@ -328,10 +329,12 @@ public class ParameterizedBuiltinCPInstruction extends ComputationCPInstruction\nString out = null;\nif( data instanceof MatrixObject ) {\nMatrixBlock matrix = (MatrixBlock) data.acquireRead();\n+ warnOnTrunction(matrix, rows, cols);\nout = DataConverter.toString(matrix, sparse, separator, lineseparator, rows, cols, decimal);\n}\nelse if( data instanceof FrameObject ) {\nFrameBlock frame = (FrameBlock) data.acquireRead();\n+ warnOnTrunction(frame, rows, cols);\nout = DataConverter.toString(frame, sparse, separator, lineseparator, rows, cols, decimal);\n}\nelse {\n@@ -344,4 +347,15 @@ public class ParameterizedBuiltinCPInstruction extends ComputationCPInstruction\nthrow new DMLRuntimeException(\"Unknown opcode : \" + opcode);\n}\n}\n+\n+ private void warnOnTrunction(CacheBlock data, int rows, int cols) {\n+ //warn on truncation because users might not be aware and use toString for verification\n+ if( (getParam(\"rows\")==null && data.getNumRows()>rows)\n+ || (getParam(\"cols\")==null && data.getNumColumns()>cols) )\n+ {\n+ LOG.warn(\"Truncating \"+data.getClass().getSimpleName()+\" of size \"\n+ + data.getNumRows()+\"x\"+data.getNumColumns()+\" to \"+rows+\"x\"+cols+\". \"\n+ + \"Use toString(X, rows=..., cols=...) if necessary.\");\n+ }\n+ }\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateRow.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateRow.java", "diff": "@@ -310,7 +310,8 @@ public class TemplateRow extends TemplateBase\nCNode cdata1 = tmp.get(hop.getInput().get(0).getHopID());\n// if one input is a matrix then we need to do vector by scalar operations\n- if(hop.getInput().get(0).getDim1() > 1 && hop.getInput().get(0).getDim2() > 1 )\n+ if(hop.getInput().get(0).getDim1() > 1 && hop.getInput().get(0).getDim2() > 1\n+ || (!hop.dimsKnown() && cdata1.getDataType()==DataType.MATRIX ) )\n{\nif( HopRewriteUtils.isUnary(hop, SUPPORTED_VECT_UNARY) ) {\nString opname = \"VECT_\"+((UnaryOp)hop).getOp().name();\n@@ -325,7 +326,6 @@ public class TemplateRow extends TemplateBase\nelse //general scalar case\n{\ncdata1 = TemplateUtils.wrapLookupIfNecessary(cdata1, hop.getInput().get(0));\n-\nString primitiveOpName = ((UnaryOp)hop).getOp().toString();\nout = new CNodeUnary(cdata1, UnaryType.valueOf(primitiveOpName));\n}\n@@ -355,7 +355,9 @@ public class TemplateRow extends TemplateBase\n// if one input is a matrix then we need to do vector by scalar operations\nif( (hop.getInput().get(0).getDim1() > 1 && hop.getInput().get(0).getDim2() > 1)\n- || (hop.getInput().get(1).getDim1() > 1 && hop.getInput().get(1).getDim2() > 1))\n+ || (hop.getInput().get(1).getDim1() > 1 && hop.getInput().get(1).getDim2() > 1)\n+ || (!(hop.dimsKnown() && hop.getInput().get(0).dimsKnown() && hop.getInput().get(1).dimsKnown())\n+ && (cdata1.getDataType().isMatrix() || cdata2.getDataType().isMatrix())))\n{\nif( HopRewriteUtils.isBinary(hop, SUPPORTED_VECT_BINARY) ) {\nif( TemplateUtils.isMatrix(cdata1) && (TemplateUtils.isMatrix(cdata2)\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateUtils.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateUtils.java", "diff": "@@ -184,7 +184,7 @@ public class TemplateUtils\npublic static RowType getRowType(Hop output, Hop... inputs) {\nHop X = inputs[0];\nHop B1 = (inputs.length>1) ? inputs[1] : null;\n- if( (X!=null && HopRewriteUtils.isEqualSize(output, X)) || X==null )\n+ if( (X!=null && HopRewriteUtils.isEqualSize(output, X)) || X==null || !X.dimsKnown() )\nreturn RowType.NO_AGG;\nelse if( ((B1!=null && output.getDim1()==X.getDim1() && output.getDim2()==B1.getDim2())\n|| (output instanceof IndexingOp && HopRewriteUtils.isColumnRangeIndexing((IndexingOp)output)))\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixDNN.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixDNN.java", "diff": "@@ -186,10 +186,10 @@ public class LibMatrixDNN {\nif(isEligibleForConv2dBackwardDataDense(params))\nStatistics.numNativeSparseConv2dBwdDataCalls.increment();\n- execute(LibMatrixDNNHelper.getConv2dBackwardDataWorkers(params), params);\n+ long nnz = execute(LibMatrixDNNHelper.getConv2dBackwardDataWorkers(params), params);\n//post-processing: maintain nnz\n- outputBlock.recomputeNonZeros();\n+ outputBlock.setNonZeros(nnz);\noutputBlock.examSparsity();\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixDNNConv2dBackwardDataHelper.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixDNNConv2dBackwardDataHelper.java", "diff": "@@ -78,22 +78,22 @@ public class LibMatrixDNNConv2dBackwardDataHelper {\nint PQ = _params.P*_params.Q; int K = _params.K; int CRS = _params.C*_params.R*_params.S;\nMatrixBlock filter = _params.input1;\nMatrixBlock dout = _params.input2;\n- MatrixBlock dout_reshaped = new MatrixBlock(PQ, K, false);\n- dout_reshaped.allocateDenseBlock();\n+ MatrixBlock outRotate = new MatrixBlock(PQ, K, dout.sparse);\n+ MatrixBlock outMM = new MatrixBlock(PQ, CRS, false);\n+ outRotate.allocateBlock();\nLibMatrixDNNRotate180Helper.Rotate180Worker rotate180Worker =\n- LibMatrixDNNRotate180Helper.Rotate180Worker.getWorker( dout, dout_reshaped, _params, true, false);\n+ LibMatrixDNNRotate180Helper.Rotate180Worker.getWorker( dout, outRotate, _params, true, false);\nlong time1 = 0; long time2 = 0;\nfor(int n = _rl; n < _ru; n++) {\n// rotate180(dout[n,]) => dout_reshaped\nrotate180Worker.execute(n, 0);\n-\n// dout_reshaped %*% filter => temp\n- MatrixBlock temp = new MatrixBlock(PQ, CRS, false);\nlong t1 = DMLScript.STATISTICS && LibMatrixDNN.DISPLAY_STATISTICS ? System.nanoTime() : 0;\n- LibMatrixDNNHelper.singleThreadedMatMult(dout_reshaped, filter, temp, true, false, _params);\n+ outMM.reset(PQ, CRS, false);\n+ LibMatrixDNNHelper.singleThreadedMatMult(outRotate, filter, outMM, !outRotate.sparse, false, _params);\nlong t2 = DMLScript.STATISTICS && LibMatrixDNN.DISPLAY_STATISTICS ? System.nanoTime() : 0;\n// col2im(temp) => output[n,]\n- LibMatrixDNNHelper.doCol2imOverSingleImage(n, temp, _params);\n+ LibMatrixDNNHelper.doCol2imOverSingleImage(n, outMM, _params);\nlong t3 = DMLScript.STATISTICS && LibMatrixDNN.DISPLAY_STATISTICS ? System.nanoTime() : 0;\nif(DMLScript.STATISTICS && LibMatrixDNN.DISPLAY_STATISTICS) {\n@@ -105,8 +105,9 @@ public class LibMatrixDNNConv2dBackwardDataHelper {\nLibMatrixDNN.loopedConvBwdDataMatMultTime.addAndGet(time1);\nLibMatrixDNN.loopedConvBwdDataCol2ImTime.addAndGet(time2);\n}\n- return 0L;\n- }\n+ //multi-threaded nnz maintenance of current working set\n+ return _params.output.recomputeNonZeros(_rl, _ru-1);\n+ }\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixDNNConv2dBackwardFilterHelper.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixDNNConv2dBackwardFilterHelper.java", "diff": "@@ -86,13 +86,12 @@ public class LibMatrixDNNConv2dBackwardFilterHelper {\nint PQ = _params.P*_params.Q, K = _params.K, CRS = _params.C*_params.R*_params.S;\nMatrixBlock dout = _params.input2;\nMatrixBlock im2ColOutBlock = new MatrixBlock(CRS, PQ, false);\n- MatrixBlock dout_reshaped = new MatrixBlock(PQ, K, dout.sparse);\n- MatrixBlock temp = new MatrixBlock(CRS, K, false);\n- dout_reshaped.allocateBlock();\n- temp.allocateBlock();\n+ MatrixBlock outRotate = new MatrixBlock(PQ, K, dout.sparse);\n+ MatrixBlock outMM = new MatrixBlock(CRS, K, false);\n+ outRotate.allocateBlock();\nIm2colWorker im2ColWorker = Im2colWorker.getWorker( _params.input1, im2ColOutBlock, _params, true, false);\n- Rotate180Worker rotate180Worker = Rotate180Worker.getWorker( dout, dout_reshaped, _params, true, false);\n+ Rotate180Worker rotate180Worker = Rotate180Worker.getWorker( dout, outRotate, _params, true, false);\ndouble [] partRet = new double[CRS*_params.K];\nlong time1 = 0; long time2 = 0;\nfor(int n = _rl; n < _ru; n++) {\n@@ -104,12 +103,12 @@ public class LibMatrixDNNConv2dBackwardFilterHelper {\nim2ColWorker.execute(n);\nlong t2 = DMLScript.STATISTICS && LibMatrixDNN.DISPLAY_STATISTICS ? System.nanoTime() : 0;\n- temp.reset(CRS, K, false);\n- LibMatrixDNNHelper.singleThreadedMatMult(im2ColOutBlock, dout_reshaped, temp, true, true, _params);\n+ outMM.reset(CRS, K, false);\n+ LibMatrixDNNHelper.singleThreadedMatMult(im2ColOutBlock, outRotate, outMM, !im2ColOutBlock.sparse, !outRotate.sparse, _params);\nlong t3 = DMLScript.STATISTICS && LibMatrixDNN.DISPLAY_STATISTICS ? System.nanoTime() : 0;\n- if( !temp.isEmptyBlock() ) //accumulate row results\n- LibMatrixMult.vectAdd(temp.getDenseBlock(), partRet, 0, 0, K*CRS);\n+ if( !outMM.isEmptyBlock() ) //accumulate row results\n+ LibMatrixMult.vectAdd(outMM.getDenseBlock(), partRet, 0, 0, K*CRS);\nif(DMLScript.STATISTICS && LibMatrixDNN.DISPLAY_STATISTICS) {\ntime1 += t2 - t1;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixDNNConv2dHelper.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixDNNConv2dHelper.java", "diff": "@@ -219,7 +219,7 @@ public class LibMatrixDNNConv2dHelper {\n// t(_im2ColOutBlock) %*% t(filter) => t(matMultOutBlock)\noutMM.reset(outMM.rlen, outMM.clen, false);\n- LibMatrixDNNHelper.singleThreadedMatMult(outIm2col, _params.input2, outMM, false, true, _params);\n+ LibMatrixDNNHelper.singleThreadedMatMult(outIm2col, _params.input2, outMM, false, false, _params);\n// Copy the matrix matMultOutBlock of shape [K X PQ] to params.output.denseBlock + destPos\npartialCopyTrans(outMM, _params.output, n*K*PQ, K, PQ);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateCell.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateCell.java", "diff": "@@ -322,10 +322,12 @@ public class TemplateCell extends TemplateBase\nprotected boolean isSparseSafe(List<Hop> roots, Hop mainInput, List<CNode> outputs, List<AggOp> aggOps, boolean onlySum) {\nboolean ret = true;\nfor( int i=0; i<outputs.size() && ret; i++ ) {\n- ret &= (HopRewriteUtils.isBinary(roots.get(i), OpOp2.MULT)\n- && roots.get(i).getInput().contains(mainInput))\n- || (HopRewriteUtils.isBinary(roots.get(i), OpOp2.DIV)\n- && roots.get(i).getInput().get(0) == mainInput)\n+ Hop root = (roots.get(i) instanceof AggUnaryOp || roots.get(i)\n+ instanceof AggBinaryOp) ? roots.get(i).getInput().get(0) : roots.get(i);\n+ ret &= (HopRewriteUtils.isBinarySparseSafe(root)\n+ && root.getInput().contains(mainInput))\n+ || (HopRewriteUtils.isBinary(root, OpOp2.DIV)\n+ && root.getInput().get(0) == mainInput)\n|| (TemplateUtils.rIsSparseSafeOnly(outputs.get(i), BinType.MULT)\n&& TemplateUtils.rContainsInput(outputs.get(i), mainInput.getHopID()));\nif( onlySum )\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/algorithms/l2-svm-predict.dml", "new_path": "scripts/algorithms/l2-svm-predict.dml", "diff": "@@ -51,6 +51,7 @@ cmdLine_Y = ifdef($Y, \" \")\ncmdLine_confusion = ifdef($confusion, \" \")\ncmdLine_accuracy = ifdef($accuracy, \" \")\ncmdLine_scores = ifdef($scores, \" \")\n+cmdLine_scoring_only = ifdef($scoring_only, FALSE)\ncmdLine_fmt = ifdef($fmt, \"text\")\nX = read($X)\n@@ -75,27 +76,30 @@ scores = b + (X %*% w[1:ncol(X),])\nif(cmdLine_scores != \" \")\nwrite(scores, cmdLine_scores, format=cmdLine_fmt)\n-if(cmdLine_Y != \" \"){\n- y = read(cmdLine_Y)\n+if(!cmdLine_scoring_only){\n+ Y = read(cmdLine_Y)\npred = (scores >= 0)\npred_labels = pred*positive_label + (1-pred)*negative_label\n- num_correct = sum(pred_labels == y)\n+ num_correct = sum(pred_labels == Y)\nacc = 100*num_correct/nrow(X)\nacc_str = \"Accuracy (%): \" + acc\nprint(acc_str)\n+\nif(cmdLine_accuracy != \" \")\nwrite(acc_str, cmdLine_accuracy)\nif(cmdLine_confusion != \" \"){\n+\npred = 2*pred - 1\n+\nif(negative_label != -1 | positive_label != +1)\n- y = 2/(positive_label - negative_label)*y - (negative_label + positive_label)/(positive_label - negative_label)\n+ Y = 2/(positive_label - negative_label)*Y - (negative_label + positive_label)/(positive_label - negative_label)\npred_is_minus = (pred == -1)\npred_is_plus = 1 - pred_is_minus\n- y_is_minus = (y == -1)\n+ y_is_minus = (Y == -1)\ny_is_plus = 1 - y_is_minus\ncheck_min_y_minus = sum(pred_is_minus*y_is_minus)\n@@ -103,21 +107,13 @@ if(cmdLine_Y != \" \"){\ncheck_max_y_minus = sum(pred_is_plus*y_is_minus)\ncheck_max_y_plus = sum(pred_is_plus*y_is_plus)\n- #s = check_min_y_minus + \",\" + check_min_y_plus\n- #s = append(s, check_max_y_minus + \",\" + check_max_y_plus)\n- #s = append(s, \"\")\n- #write(s, cmdLine_confusion)\n-\n- confusion_mat = matrix(0, rows=3, cols=3)\n- confusion_mat[1,2] = negative_label\n- confusion_mat[1,3] = positive_label\n- confusion_mat[2,1] = negative_label\n- confusion_mat[3,1] = positive_label\n- confusion_mat[2,2] = check_min_y_minus\n- confusion_mat[2,3] = check_max_y_minus\n- confusion_mat[3,2] = check_min_y_plus\n- confusion_mat[3,3] = check_max_y_plus\n+ confusion_mat = matrix(0, rows=2, cols=2)\n+ confusion_mat[1,1] = check_min_y_minus\n+ confusion_mat[1,2] = check_min_y_plus\n+ confusion_mat[2,1] = check_max_y_minus\n+ confusion_mat[2,2] = check_max_y_plus\nwrite(confusion_mat, cmdLine_confusion, format=\"csv\")\n}\n}\n+\n" }, { "change_type": "MODIFY", "old_path": "scripts/algorithms/l2-svm.dml", "new_path": "scripts/algorithms/l2-svm.dml", "diff": "@@ -80,8 +80,6 @@ if(check_min != -1 | check_max != +1)\npositive_label = check_max\nnegative_label = check_min\n-continue = 1\n-\nintercept = cmdLine_icpt\nif(intercept != 0 & intercept != 1)\nstop(\"Stopping due to invalid argument: Currently supported intercept options are 0 and 1\")\n@@ -118,14 +116,16 @@ s = g_old\nXw = matrix(0, rows=nrow(X), cols=1)\ndebug_str = \"# Iter, Obj\"\niter = 0\n-while(continue == 1 & iter < maxiterations) {\n+continue = TRUE\n+while(continue & iter < maxiterations) {\n# minimizing primal obj along direction s\nstep_sz = 0\nXd = X %*% s\nwd = lambda * sum(w * s)\ndd = lambda * sum(s * s)\n- continue1 = 1\n- while(continue1 == 1){\n+\n+ continue1 = TRUE\n+ while(continue1){\ntmp_Xw = Xw + step_sz*Xd\nout = 1 - Y * (tmp_Xw)\nsv = (out > 0)\n@@ -133,9 +133,8 @@ while(continue == 1 & iter < maxiterations) {\ng = wd + step_sz*dd - sum(out * Y * Xd)\nh = dd + sum(Xd * sv * Xd)\nstep_sz = step_sz - g/h\n- if (g*g/h < 0.0000000001){\n- continue1 = 0\n- }\n+\n+ continue1 = (gg/h >= 0.0000000001);\n}\n#update weights\n@@ -152,19 +151,13 @@ while(continue == 1 & iter < maxiterations) {\ndebug_str = append(debug_str, iter + \",\" + obj)\ntmp = sum(s * g_old)\n- if(step_sz*tmp < epsilon*obj){\n- continue = 0\n- }\n+ continue = (step_sz*tmp >= epsilon*obj & sum(s^2) != 0);\n#non-linear CG step\nbe = sum(g_new * g_new)/sum(g_old * g_old)\ns = be * s + g_new\ng_old = g_new\n- if(sum(s^2) == 0){\n- continue = 0\n- }\n-\niter = iter + 1\n}\n@@ -174,8 +167,11 @@ extra_model_params[2,1] = negative_label\nextra_model_params[3,1] = intercept\nextra_model_params[4,1] = dimensions\n+weights = w\nw = t(cbind(t(w), t(extra_model_params)))\nwrite(w, $model, format=cmdLine_fmt)\n+# write(extra_model_params, \" \", format=cmdLine_fmt)\n+# write(weights, \" \", format=cmdLine_fmt)\nlogFile = $Log\nif(logFile != \" \") {\n" }, { "change_type": "MODIFY", "old_path": "scripts/algorithms/m-svm-predict.dml", "new_path": "scripts/algorithms/m-svm-predict.dml", "diff": "# accuracy (%) for the predictions\n#\n# Example Usage:\n-# hadoop jar SystemML.jar -f m-svm-predict.dml -nvargs X=data Y=labels model=model scores=scores accuracy=accuracy confusion=confusion fmt=\"text\"\n+# hadoop jar SystemML.jar -f m-svm-predict.dml -nvargs X=data Y=labels scoring_only=FALSE model=model scores=scores accuracy=accuracy confusion=confusion fmt=\"text\"\n#\ncmdLine_Y = ifdef($Y, \" \")\ncmdLine_confusion = ifdef($confusion, \" \")\ncmdLine_accuracy = ifdef($accuracy, \" \")\ncmdLine_scores = ifdef($scores, \" \")\n+cmdLine_scoring_only = ifdef($scoring_only, FALSE)\ncmdLine_fmt = ifdef($fmt, \"text\")\nX = read($X);\n@@ -59,26 +60,26 @@ scores = X %*% W[1:m,] + ones %*% b;\nif(cmdLine_scores != \" \")\nwrite(scores, cmdLine_scores, format=cmdLine_fmt);\n-if(cmdLine_Y != \" \"){\n- y = read(cmdLine_Y);\n+if(!cmdLine_scoring_only){\n+ Y = read(cmdLine_Y);\n- if(min(y) < 1)\n+ if(min(Y) < 1)\nstop(\"Stopping due to invalid argument: Label vector (Y) must be recoded\")\npred = rowIndexMax(scores);\n- correct_percentage = sum((pred - y) == 0) / N * 100;\n+ correct_percentage = sum((pred - Y) == 0) / N * 100;\nacc_str = \"Accuracy (%): \" + correct_percentage\nprint(acc_str)\nif(cmdLine_accuracy != \" \")\nwrite(acc_str, cmdLine_accuracy)\n- num_classes_ground_truth = max(y)\n+ num_classes_ground_truth = max(Y)\nif(num_classes < num_classes_ground_truth)\nnum_classes = num_classes_ground_truth\nif(cmdLine_confusion != \" \"){\n- confusion_mat = table(y, pred, num_classes, num_classes)\n+ confusion_mat = table(Y, pred, num_classes, num_classes)\nwrite(confusion_mat, cmdLine_confusion, format=\"csv\")\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "scripts/algorithms/m-svm.dml", "new_path": "scripts/algorithms/m-svm.dml", "diff": "# Assume input and output directories are on hdfs as INPUT_DIR and OUTPUT_DIR\n# Assume epsilon = 0.001, lambda=1.0, max_iterations = 100\n#\n+# INPUT PARAMETERS:\n+# ---------------------------------------------------------------------------------------------\n+# NAME TYPE DEFAULT MEANING\n+# ---------------------------------------------------------------------------------------------\n+# X String --- Location to read the matrix X of feature vectors\n+# Y String --- Location to read response matrix Y\n+# icpt Int 0 Intercept presence\n+# 0 = no intercept\n+# 1 = add intercept;\n+# tol Double 0.001 Tolerance (epsilon);\n+# reg Double 1.0 Regularization parameter\n+# maxiter Int 100 Maximum number of conjugate gradient iterations\n+# model String --- Location to write model\n+# fmt String \"text\" The output format of the output, such as \"text\" or \"csv\"\n+# Log String --- [OPTIONAL] Location to write the log file\n+# ---------------------------------------------------------------------------------------------\n+#\n# hadoop jar SystemML.jar -f $SVM_HOME/m-svm.dml -nvargs X=$INPUT_DIR/X Y=$INPUT_DIR/y icpt=intercept tol=.001 reg=1.0 maxiter=100 model=$OUTPUT_DIR/w Log=$OUTPUT_DIR/Log fmt=\"text\"\n#\n@@ -57,9 +74,11 @@ if(intercept != 0 & intercept != 1)\nmin_y = min(Y)\nif(min_y < 1)\nstop(\"Stopping due to invalid argument: Label vector (Y) must be recoded\")\n+\nnum_classes = max(Y)\nif(num_classes == 1)\nstop(\"Stopping due to invalid argument: Maximum label value is 1, need more than one class to learn a multi-class classifier\")\n+\nmod1 = Y %% 1\nmod1_should_be_nrow = sum(abs(mod1 == 0))\nif(mod1_should_be_nrow != nrow(Y))\n@@ -92,9 +111,11 @@ if(intercept == 1){\nw = matrix(0, rows=num_rows_in_w, cols=num_classes)\ndebug_mat = matrix(-1, rows=max_iterations, cols=num_classes)\n+\nparfor(iter_class in 1:num_classes){\nY_local = 2 * (Y == iter_class) - 1\nw_class = matrix(0, rows=num_features, cols=1)\n+\nif (intercept == 1) {\nzero_matrix = matrix(0, rows=1, cols=1);\nw_class = t(cbind(t(w_class), zero_matrix));\n@@ -105,15 +126,16 @@ parfor(iter_class in 1:num_classes){\nXw = matrix(0, rows=nrow(X), cols=1)\niter = 0\n- continue = 1\n- while(continue == 1) {\n+ continue = TRUE\n+ while(continue & iter < maxiterations) {\n# minimizing primal obj along direction s\nstep_sz = 0\nXd = X %*% s\nwd = lambda * sum(w_class * s)\ndd = lambda * sum(s * s)\n- continue1 = 1\n- while(continue1 == 1){\n+\n+ continue1 = TRUE\n+ while(continue1){\ntmp_Xw = Xw + step_sz*Xd\nout = 1 - Y_local * (tmp_Xw)\nsv = (out > 0)\n@@ -121,9 +143,9 @@ parfor(iter_class in 1:num_classes){\ng = wd + step_sz*dd - sum(out * Y_local * Xd)\nh = dd + sum(Xd * sv * Xd)\nstep_sz = step_sz - g/h\n- if (g*g/h < 0.0000000001){\n- continue1 = 0\n- }\n+\n+ continue1 = (g*g/h >= 0.0000000001)\n+\n}\n#update weights\n@@ -142,30 +164,27 @@ parfor(iter_class in 1:num_classes){\nprint(\"For class \" + iter_class + \" iteration \" + iter + \" training accuracy: \" + train_acc)\ndebug_mat[iter+1,iter_class] = obj\n- if((step_sz*tmp < epsilon*obj) | (iter >= max_iterations-1)){\n- continue = 0\n- }\n+ continue = (step_sz*tmp >= epsilon*obj & sum(s^2) != 0);\n#non-linear CG step\nbe = sum(g_new * g_new)/sum(g_old * g_old)\ns = be * s + g_new\ng_old = g_new\n- if(sum(s^2) == 0){\n- continue = 0\n- }\n-\niter = iter + 1\n}\nw[,iter_class] = w_class\n-}\n+} # parfor loop\nextra_model_params = matrix(0, rows=2, cols=ncol(w))\nextra_model_params[1, 1] = intercept\nextra_model_params[2, 1] = dimensions\n+weights = w\nw = t(cbind(t(w), t(extra_model_params)))\nwrite(w, $model, format=cmdLine_fmt)\n+# write(extra_model_params, \" \", format=cmdLine_fmt)\n+# write(weights, \" \", format=cmdLine_fmt)\ndebug_str = \"# Class, Iter, Obj\"\nfor(iter_class in 1:ncol(debug_mat)){\n@@ -175,7 +194,8 @@ for(iter_class in 1:ncol(debug_mat)){\ndebug_str = append(debug_str, iter_class + \",\" + iter + \",\" + obj)\n}\n}\n+\nlogFile = $Log\n-if(logFile != \" \") {\n+if(logFile != \" \")\nwrite(debug_str, logFile)\n-}\n\\ No newline at end of file\n+\n" } ]

[ { "change_type": "MODIFY", "old_path": "scripts/algorithms/l2-svm.dml", "new_path": "scripts/algorithms/l2-svm.dml", "diff": "@@ -134,7 +134,7 @@ while(continue & iter < maxiterations) {\nh = dd + sum(Xd * sv * Xd)\nstep_sz = step_sz - g/h\n- continue1 = (gg/h >= 0.0000000001);\n+ continue1 = (g*g/h >= 0.0000000001);\n}\n#update weights\n@@ -151,14 +151,15 @@ while(continue & iter < maxiterations) {\ndebug_str = append(debug_str, iter + \",\" + obj)\ntmp = sum(s * g_old)\n- continue = (step_sz*tmp >= epsilon*obj & sum(s^2) != 0);\n#non-linear CG step\nbe = sum(g_new * g_new)/sum(g_old * g_old)\ns = be * s + g_new\ng_old = g_new\n+ continue = (step_sz*tmp >= epsilon*obj & sum(s^2) != 0);\niter = iter + 1\n+\n}\nextra_model_params = matrix(0, rows=4, cols=1)\n@@ -167,11 +168,8 @@ extra_model_params[2,1] = negative_label\nextra_model_params[3,1] = intercept\nextra_model_params[4,1] = dimensions\n-weights = w\nw = t(cbind(t(w), t(extra_model_params)))\nwrite(w, $model, format=cmdLine_fmt)\n-# write(extra_model_params, \" \", format=cmdLine_fmt)\n-# write(weights, \" \", format=cmdLine_fmt)\nlogFile = $Log\nif(logFile != \" \") {\n" }, { "change_type": "MODIFY", "old_path": "scripts/algorithms/m-svm.dml", "new_path": "scripts/algorithms/m-svm.dml", "diff": "# Example Usage:\n# Assume SVM_HOME is set to the home of the dml script\n# Assume input and output directories are on hdfs as INPUT_DIR and OUTPUT_DIR\n-# Assume epsilon = 0.001, lambda=1.0, max_iterations = 100\n+# Assume epsilon = 0.001, lambda=1.0, maxiterations = 100\n#\n# INPUT PARAMETERS:\n# ---------------------------------------------------------------------------------------------\n@@ -92,8 +92,8 @@ lambda = cmdLine_reg\nif(lambda < 0)\nstop(\"Stopping due to invalid argument: Regularization constant (reg) must be non-negative\")\n-max_iterations = cmdLine_maxiter\n-if(max_iterations < 1)\n+maxiterations = cmdLine_maxiter\n+if(maxiterations < 1)\nstop(\"Stopping due to invalid argument: Maximum iterations should be a positive integer\")\nnum_samples = nrow(X)\n@@ -110,7 +110,7 @@ if(intercept == 1){\n}\nw = matrix(0, rows=num_rows_in_w, cols=num_classes)\n-debug_mat = matrix(-1, rows=max_iterations, cols=num_classes)\n+debug_mat = matrix(-1, rows=maxiterations, cols=num_classes)\nparfor(iter_class in 1:num_classes){\nY_local = 2 * (Y == iter_class) - 1\n@@ -145,7 +145,6 @@ parfor(iter_class in 1:num_classes){\nstep_sz = step_sz - g/h\ncontinue1 = (g*g/h >= 0.0000000001)\n-\n}\n#update weights\n@@ -164,13 +163,12 @@ parfor(iter_class in 1:num_classes){\nprint(\"For class \" + iter_class + \" iteration \" + iter + \" training accuracy: \" + train_acc)\ndebug_mat[iter+1,iter_class] = obj\n- continue = (step_sz*tmp >= epsilon*obj & sum(s^2) != 0);\n-\n#non-linear CG step\nbe = sum(g_new * g_new)/sum(g_old * g_old)\ns = be * s + g_new\ng_old = g_new\n+ continue = (step_sz*tmp >= epsilon*obj & sum(s^2) != 0);\niter = iter + 1\n}\n@@ -180,11 +178,9 @@ parfor(iter_class in 1:num_classes){\nextra_model_params = matrix(0, rows=2, cols=ncol(w))\nextra_model_params[1, 1] = intercept\nextra_model_params[2, 1] = dimensions\n-weights = w\n+\nw = t(cbind(t(w), t(extra_model_params)))\nwrite(w, $model, format=cmdLine_fmt)\n-# write(extra_model_params, \" \", format=cmdLine_fmt)\n-# write(weights, \" \", format=cmdLine_fmt)\ndebug_str = \"# Class, Iter, Obj\"\nfor(iter_class in 1:ncol(debug_mat)){\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateRow.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateRow.java", "diff": "@@ -362,6 +362,7 @@ public class TemplateRow extends TemplateBase\nif( (hop.getInput().get(0).getDim1() > 1 && hop.getInput().get(0).getDim2() > 1)\n|| (hop.getInput().get(1).getDim1() > 1 && hop.getInput().get(1).getDim2() > 1)\n|| (!(hop.dimsKnown() && hop.getInput().get(0).dimsKnown() && hop.getInput().get(1).dimsKnown())\n+ && (hop.getDim2() != 1) //not a known vector output\n&& (cdata1.getDataType().isMatrix() || cdata2.getDataType().isMatrix())))\n{\nif( HopRewriteUtils.isBinary(hop, SUPPORTED_VECT_BINARY) ) {\n" } ]

[ { "change_type": "MODIFY", "old_path": "docs/beginners-guide-caffe2dml.md", "new_path": "docs/beginners-guide-caffe2dml.md", "diff": "@@ -64,22 +64,27 @@ lenet.summary()\nOutput:\n```\n-+-----+---------------+--------------+------------+---------+-----------+---------+\n-| Name| Type| Output| Weight| Bias| Top| Bottom|\n-+-----+---------------+--------------+------------+---------+-----------+---------+\n-|mnist| Data| (, 1, 28, 28)| | |mnist,mnist| |\n-|conv1| Convolution|(, 32, 28, 28)| [32 X 25]| [32 X 1]| conv1| mnist|\n-|relu1| ReLU|(, 32, 28, 28)| | | relu1| conv1|\n-|pool1| Pooling|(, 32, 14, 14)| | | pool1| relu1|\n-|conv2| Convolution|(, 64, 14, 14)| [64 X 800]| [64 X 1]| conv2| pool1|\n-|relu2| ReLU|(, 64, 14, 14)| | | relu2| conv2|\n-|pool2| Pooling| (, 64, 7, 7)| | | pool2| relu2|\n-| ip1| InnerProduct| (, 512, 1, 1)|[3136 X 512]|[1 X 512]| ip1| pool2|\n-|relu3| ReLU| (, 512, 1, 1)| | | relu3| ip1|\n-|drop1| Dropout| (, 512, 1, 1)| | | drop1| relu3|\n-| ip2| InnerProduct| (, 10, 1, 1)| [512 X 10]| [1 X 10]| ip2| drop1|\n-| loss|SoftmaxWithLoss| (, 10, 1, 1)| | | loss|ip2,mnist|\n-+-----+---------------+--------------+------------+---------+-----------+---------+\n++-----+---------------+--------------+------------+---------+-----------+---------+--------------------+\n+| Name| Type| Output| Weight| Bias| Top| Bottom|Memory* (train/test)|\n++-----+---------------+--------------+------------+---------+-----------+---------+--------------------+\n+|mnist| Data| (, 1, 28, 28)| | |mnist,mnist| | 1/0|\n+|conv1| Convolution|(, 32, 28, 28)| [32 X 25]| [32 X 1]| conv1| mnist| 25/12|\n+|relu1| ReLU|(, 32, 28, 28)| | | relu1| conv1| 25/12|\n+|pool1| Pooling|(, 32, 14, 14)| | | pool1| relu1| 6/3|\n+|conv2| Convolution|(, 64, 14, 14)| [64 X 800]| [64 X 1]| conv2| pool1| 38/7|\n+|relu2| ReLU|(, 64, 14, 14)| | | relu2| conv2| 12/6|\n+|pool2| Pooling| (, 64, 7, 7)| | | pool2| relu2| 3/2|\n+| ip1| InnerProduct| (, 512, 1, 1)|[3136 X 512]|[1 X 512]| ip1| pool2| 797/13|\n+|relu3| ReLU| (, 512, 1, 1)| | | relu3| ip1| 1/0|\n+|drop1| Dropout| (, 512, 1, 1)| | | drop1| relu3| 1/0|\n+| ip2| InnerProduct| (, 10, 1, 1)| [512 X 10]| [1 X 10]| ip2| drop1| 3/0|\n+| loss|SoftmaxWithLoss| (, 10, 1, 1)| | | loss|ip2,mnist| 0/0|\n++-----+---------------+--------------+------------+---------+-----------+---------+--------------------+\n+\n+Total number of layer outputs/errors/weights/bias/gradients: 5568768/5568768/1662752/618/106455680\n+Total memory requirements for parameters* for train/test: 910/55\n+[Advanced] Key network statistics to compute intermediate CP overhead batchSize/maxThreads/1-thread im2col*(sum, max)/1-thread reshape_col*(sum, max): 64/48/(1, 1)/(0, 0).\n+* => memory in megabytes assuming the parameters are in double precision and in dense format.\n```\nTo train the above lenet model, we use the MNIST dataset.\n" }, { "change_type": "MODIFY", "old_path": "src/main/scala/org/apache/sysml/api/dl/Caffe2DML.scala", "new_path": "src/main/scala/org/apache/sysml/api/dl/Caffe2DML.scala", "diff": "@@ -50,6 +50,8 @@ import java.util.Random\nimport org.apache.commons.logging.Log\nimport org.apache.commons.logging.LogFactory\nimport org.apache.sysml.runtime.controlprogram.parfor.stat.InfrastructureAnalyzer\n+import org.apache.sysml.hops.OptimizerUtils\n+import java.lang.Double\n/***************************************************************************************\nDESIGN OF CAFFE2DML:\n@@ -306,10 +308,21 @@ class Caffe2DML(val sc: SparkContext,\ndef getTrainAlgo(): String = if (inputs.containsKey(\"$train_algo\")) inputs.get(\"$train_algo\") else \"minibatch\"\ndef getTestAlgo(): String = if (inputs.containsKey(\"$test_algo\")) inputs.get(\"$test_algo\") else \"minibatch\"\n+ private def getMemInBytes(l:CaffeLayer, batchSize:Int, isTraining:Boolean):Long = {\n+ val numLayerOutput = l.outputShape._1.toLong * l.outputShape._2.toLong * l.outputShape._3.toLong * batchSize\n+ val numLayerError = numLayerOutput\n+ val numLayerWeights = if(l.weightShape != null) l.weightShape()(0).toLong * l.weightShape()(1).toLong else 0\n+ val numLayerBias = if(l.biasShape != null)l.biasShape()(0).toLong * l.biasShape()(1).toLong else 0\n+ val numLayerGradients = (numLayerWeights + numLayerBias) * batchSize\n+ if(isTraining) (numLayerOutput + numLayerError + numLayerWeights + numLayerBias + numLayerGradients)*Double.BYTES\n+ else (numLayerOutput + numLayerWeights + numLayerBias)*Double.BYTES\n+ }\ndef summary(sparkSession: org.apache.spark.sql.SparkSession): Unit = {\n- val header = Seq(\"Name\", \"Type\", \"Output\", \"Weight\", \"Bias\", \"Top\", \"Bottom\")\n- val entries = net.getLayers\n- .map(l => (l, net.getCaffeLayer(l)))\n+ val layers = net.getLayers .map(l => (l, net.getCaffeLayer(l)))\n+ val numDataLayers = layers.filter(l => l._2.isInstanceOf[Data]).length\n+ val batchSize = if(numDataLayers == 1) layers.filter(l => l._2.isInstanceOf[Data]).map(l => l._2.param.getDataParam.getBatchSize).get(0) else -1\n+ val header = Seq(\"Name\", \"Type\", \"Output\", \"Weight\", \"Bias\", \"Top\", \"Bottom\", \"Memory* (train/test)\")\n+ val entries = layers\n.map(l => {\nval layer = l._2\n(l._1,\n@@ -318,10 +331,35 @@ class Caffe2DML(val sc: SparkContext,\nif (layer.weightShape != null) \"[\" + layer.weightShape()(0) + \" X \" + layer.weightShape()(1) + \"]\" else \"\",\nif (layer.biasShape != null) \"[\" + layer.biasShape()(0) + \" X \" + layer.biasShape()(1) + \"]\" else \"\",\nlayer.param.getTopList.mkString(\",\"),\n- layer.param.getBottomList.mkString(\",\"))\n+ layer.param.getBottomList.mkString(\",\"),\n+ OptimizerUtils.toMB(getMemInBytes(l._2, batchSize, true)) + \"/\" + OptimizerUtils.toMB(getMemInBytes(l._2, batchSize, false))\n+ )\n})\nimport sparkSession.implicits._\nsc.parallelize(entries).toDF(header: _*).show(net.getLayers.size)\n+\n+ val numLayerOutput = layers.map(l => l._2.outputShape._1.toLong * l._2.outputShape._2.toLong * l._2.outputShape._3.toLong).sum * batchSize\n+ val numLayerError = numLayerOutput\n+ val numLayerWeights = layers.map(l => if(l._2.weightShape != null) l._2.weightShape()(0).toLong * l._2.weightShape()(1).toLong else 0).sum\n+ val numLayerBias = layers.map(l => if(l._2.biasShape != null) l._2.biasShape()(0).toLong * l._2.biasShape()(1).toLong else 0).sum\n+ val numLayerGradients = (numLayerWeights + numLayerBias) * batchSize\n+ val convLayers = layers.filter(l => l._2.isInstanceOf[Convolution]).map(l => l._2.asInstanceOf[Convolution])\n+ val crspq = convLayers.map(l => l.numChannels.toLong*l.kernel_h.toLong*l.kernel_w.toLong*l.outputShape._2.toLong*l.outputShape._3.toLong)\n+ val kpq = convLayers.map(l => l.outputShape._1.toLong*l.outputShape._2.toLong*l.outputShape._3.toLong)\n+\n+ if(getTrainAlgo().equals(\"minibatch\") && getTestAlgo().equals(\"minibatch\")) {\n+ System.out.println(\"Total number of layer outputs/errors/weights/bias/gradients: \" + numLayerOutput + \"/\" + numLayerError +\n+ \"/\" + numLayerWeights + \"/\" + numLayerBias + \"/\" + numLayerGradients)\n+ System.out.println(\"Total memory requirements for parameters* for train/test: \" +\n+ OptimizerUtils.toMB(layers.map(l => getMemInBytes(l._2, batchSize, true)).sum) + \"/\" +\n+ OptimizerUtils.toMB(layers.map(l => getMemInBytes(l._2, batchSize, false)).sum))\n+ System.out.println(\"[Advanced] Key network statistics to compute intermediate CP overhead \" +\n+ \"batchSize/maxThreads/1-thread im2col*(sum, max)/1-thread reshape_col*(sum, max): \" +\n+ batchSize + \"/\" + OptimizerUtils.getConstrainedNumThreads(-1) + \"/(\" +\n+ OptimizerUtils.toMB(crspq.sum*Double.BYTES) + \", \" + OptimizerUtils.toMB(crspq.max*Double.BYTES) + \")/(\" +\n+ OptimizerUtils.toMB(kpq.sum*Double.BYTES) + \", \" + OptimizerUtils.toMB(kpq.max*Double.BYTES) + \").\")\n+ }\n+ System.out.println(\"* => memory in megabytes assuming the parameters are in double precision and in dense format.\")\n}\n// ================================================================================================\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/context/ExecutionConfig.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/context/ExecutionConfig.java", "diff": "@@ -69,38 +69,14 @@ public class ExecutionConfig {\n* @throws DMLRuntimeException if DMLRuntimeException occurs\n*/\npublic static ExecutionConfig getConfigForSimpleVectorOperations(int numCells) throws DMLRuntimeException {\n+ if(numCells == 0)\n+ throw new DMLRuntimeException(\"Attempting to invoke a kernel with 0 threads\");\nint deviceNumber = 0;\nint blockDimX = getMaxBlockDim(deviceNumber);\nint gridDimX = (int) Math.ceil((double) numCells / blockDimX);\nreturn new ExecutionConfig(gridDimX, blockDimX);\n}\n- /**\n- * Use this for simple matrix operations and use following in the kernel\n- * <code>\n- * int ix = blockIdx.x * blockDim.x + threadIdx.x;\n- * int iy = blockIdx.y * blockDim.y + threadIdx.y;\n- * </code>\n- * <p>\n- * This tries to schedule as minimum grids as possible.\n- *\n- * @param rlen number of rows\n- * @param clen number of columns\n- * @return execution configuration\n- * @throws DMLRuntimeException if DMLRuntimeException occurs\n- */\n- public static ExecutionConfig getConfigForMatrixOperations(int rlen, int clen) throws DMLRuntimeException {\n- int deviceNumber = 0;\n- int maxBlockDim = getMaxBlockDim(deviceNumber);\n- int blockDimX = (int) Math.min(maxBlockDim, rlen);\n- int gridDimX = (int) Math.ceil((double) rlen / blockDimX);\n- int blockDimY = (int) Math.min(Math.floor(((double) maxBlockDim) / blockDimX), clen);\n- int gridDimY = (int) Math.ceil((double) clen / blockDimY);\n- if (gridDimY > 65535)\n- throw new DMLRuntimeException(\"Internal Error: gridDimY must be less than 65535 for all supported CUDA compute capabilites!\");\n- return new ExecutionConfig(gridDimX, gridDimY, blockDimX, blockDimY);\n- }\n-\n/**\n* Use this for simple vector operations and use following in the kernel\n* <code>\n@@ -116,7 +92,6 @@ public class ExecutionConfig {\nreturn getConfigForSimpleVectorOperations(rlen * clen);\n}\n-\npublic ExecutionConfig(int gridDimX, int blockDimX) {\nthis.gridDimX = gridDimX;\nthis.blockDimX = blockDimX;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixCUDA.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixCUDA.java", "diff": "@@ -1821,17 +1821,19 @@ public class LibMatrixCUDA {\n*/\nprotected static void sliceSparseDense(GPUContext gCtx, String instName, CSRPointer inPointer, Pointer outPointer,\nint rl, int ru, int cl, int cu, int inClen) throws DMLRuntimeException {\n+ int size = getNnz(inPointer, rl, ru);\n+ // Return since nnz of the output is 0 as outPointer is expected to be zeroed out.\n+ if(size == 0) return;\n+\nint retRlen = ru - rl + 1;\nlong t0 = GPUStatistics.DISPLAY_STATISTICS ? System.nanoTime() : 0;\nint retClen = cu - cl + 1;\n- int size = -1; String kernel = null; String timer = null;\n-\n+ String kernel = null; String timer = null;\n// Note: row-wise parallelization scheme iterates over input rows in single thread\n// whereas nnz parallelization scheme iterates over number of output rows in single thread.\nif(inClen > 10 && retClen > 2*retRlen) {\n// Perform nnz parallelization for wide and short matrices\n- size = getNnz(inPointer, rl, ru);\ntimer = GPUInstruction.MISC_TIMER_RIX_SPARSE_DENSE_OP_NNZ;\nkernel = \"slice_sparse_dense_nnz\";\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/gpu/GPUTests.java", "new_path": "src/test/java/org/apache/sysml/test/gpu/GPUTests.java", "diff": "@@ -55,7 +55,7 @@ public abstract class GPUTests extends AutomatedTestBase {\nprivate static final boolean PRINT_MAT_ERROR = false;\n// We will use this flag until lower precision is supported on CP.\n- private final static String DATA_TYPE = \"double\";\n+ private final static String FLOATING_POINT_PRECISION = \"double\";\nprotected final double SINGLE_PRECISION_THRESHOLD = 1e-3; // for relative error\n@@ -75,9 +75,9 @@ public abstract class GPUTests extends AutomatedTestBase {\n* @return a valid threshold\n*/\nprotected double getTHRESHOLD() {\n- if(DATA_TYPE.equals(\"double\")) return DOUBLE_PRECISION_THRESHOLD;\n- else if(DATA_TYPE.equals(\"float\")) return SINGLE_PRECISION_THRESHOLD;\n- else throw new RuntimeException(\"Unsupported datatype:\" + DATA_TYPE);\n+ if(FLOATING_POINT_PRECISION.equals(\"double\")) return DOUBLE_PRECISION_THRESHOLD;\n+ else if(FLOATING_POINT_PRECISION.equals(\"single\")) return SINGLE_PRECISION_THRESHOLD;\n+ else throw new RuntimeException(\"Unsupported precision:\" + FLOATING_POINT_PRECISION);\n}\n@After\n@@ -263,7 +263,7 @@ public abstract class GPUTests extends AutomatedTestBase {\nformat.format(\n\"Relative error(%f) is more than threshold (%f). Expected = %f, Actual = %f, differed at [%d, %d]\",\nrelativeError, getTHRESHOLD(), expectedDouble, actualDouble, i, j);\n- if(DATA_TYPE.equals(\"double\"))\n+ if(FLOATING_POINT_PRECISION.equals(\"double\"))\nAssert.assertTrue(format.toString(), relativeError < getTHRESHOLD());\nelse\nAssert.assertTrue(format.toString(), relativeError < getTHRESHOLD() || absoluteError < getTHRESHOLD());\n@@ -324,7 +324,7 @@ public abstract class GPUTests extends AutomatedTestBase {\nprotected List<Object> runOnGPU(SparkSession spark, String scriptStr, Map<String, Object> inputs,\nList<String> outStrs) {\nMLContext gpuMLC = new MLContext(spark);\n- gpuMLC.setConfigProperty(\"sysml.gpu.dataType\", DATA_TYPE);\n+ gpuMLC.setConfigProperty(\"sysml.floating.point.precision\", FLOATING_POINT_PRECISION);\ngpuMLC.setGPU(true);\ngpuMLC.setForceGPU(true);\ngpuMLC.setStatistics(true);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/ConvolutionOp.java", "new_path": "src/main/java/org/apache/sysml/hops/ConvolutionOp.java", "diff": "@@ -47,14 +47,23 @@ public class ConvolutionOp extends Hop implements MultiThreadedHop\nprivate static final boolean THROW_ERROR_IF_INFERRED_SHAPE_MISMATCH = true;\n// -------------------------------------------------------------------------\n+ // Specifies the type of this hop\nprivate Hop.ConvOp op;\n-\nprivate int _maxNumThreads = -1; //-1 for unlimited\nprivate ConvolutionOp() {\n//default constructor for clone\n}\n+ /**\n+ * Create a hop from the builtin expression\n+ *\n+ * @param l name of the hop\n+ * @param dt datatype (only supports matrix datatype)\n+ * @param vt valuetype (only supports matrix valuetype)\n+ * @param o type of this hop\n+ * @param inp input hops\n+ */\npublic ConvolutionOp(String l, DataType dt, ValueType vt, ConvOp o, ArrayList<Hop> inp)\n{\nsuper(l, dt, vt);\n@@ -75,8 +84,7 @@ public class ConvolutionOp extends Hop implements MultiThreadedHop\nHopsException.check(_input.size() >= 1, this, \"should have at least one input but has %d inputs\", _input.size());\n}\n- public ConvOp getOp()\n- {\n+ public ConvOp getOp() {\nreturn op;\n}\n@@ -163,77 +171,129 @@ public class ConvolutionOp extends Hop implements MultiThreadedHop\nreturn input instanceof ConvolutionOp && ((ConvolutionOp) input).getOp() == ConvOp.DIRECT_CONV2D;\n}\n+ /**\n+ * Compares the input parameters for max_pool/max_pool_backward operations\n+ *\n+ * @return true if the following parameters match: stride=[stride, stride], padding=[pad, pad], input_shape=[numImg, numChannels, imgSize, imgSize], pool_size=[poolSize1, poolSize2]\n+ */\n+ private static boolean isPoolingParametersEqualAndKnown(ConvolutionParameters param1, ConvolutionParameters param2) {\n+ return isEqualAndKnown(param1.stride_h, param2.stride_h) && isEqualAndKnown(param1.stride_w, param2.stride_w) &&\n+ isEqualAndKnown(param1.pad_h, param2.pad_h) && isEqualAndKnown(param1.pad_w, param2.pad_w) &&\n+ isEqualAndKnown(param1.R, param2.R) && isEqualAndKnown(param1.S, param2.S) &&\n+ isEqualAndKnown(param1.N, param2.N) && isEqualAndKnown(param1.C, param2.C) &&\n+ isEqualAndKnown(param1.H, param2.H) && isEqualAndKnown(param1.W, param2.W);\n+ }\n+\n+ private static boolean isEqualAndKnown(int val1, int val2) {\n+ return val1 >= 0 && val2 >= 0 && val1 == val2;\n+ }\n+\n+ /**\n+ * Returns the output lop of maxpool operation with same parameters as this hop.\n+ * If corresponding output lop is not found or if this is not a max_pool_backward operation, this function returns null\n+ *\n+ * @return output lop of maxpool operation with same parameters as this hop\n+ * @throws HopsException if error\n+ * @throws LopsException if error\n+ */\n+ private Lop getMaxPoolOutputLop() throws HopsException, LopsException {\n+ if(op != ConvOp.MAX_POOLING_BACKWARD)\n+ return null;\n+\n+ Hop inputImage = getInput().get(0);\n+ for(Hop tmpParent : inputImage.getParent()) {\n+ if(!(tmpParent instanceof ConvolutionOp))\n+ continue;\n+ ConvolutionOp parent = (ConvolutionOp) tmpParent;\n+ if(parent.getOp() == ConvOp.MAX_POOLING && isPoolingParametersEqualAndKnown(parent._cachedParams, _cachedParams)) {\n+ return parent.constructLops();\n+ }\n+ }\n+ return null;\n+ }\n+\npublic Lop constructConvolutionLops(ExecType et, ArrayList<Hop> inputs) throws HopsException, LopsException {\nif(inputs.size() != getNumExpectedInputs())\nthrow new HopsException(\"Incorrect number of inputs for \" + op.name());\n- Lop in = null; Lop in2 = null;\n- ArrayList<Hop> inputs1 = inputs;\n- int k = OptimizerUtils.getConstrainedNumThreads(_maxNumThreads);\n+ // ---------------------------------------------------------------\n+ // Deal with fused operators and contruct lhsInputLop/optionalRhsInputLop\n+ Lop lhsInputLop = null; Lop optionalRhsInputLop = null;\n+ ArrayList<Hop> inputsOfPotentiallyFusedOp = inputs;\nOperationTypes lopOp = HopsConv2Lops.get(op);\n// RELU_MAX_POOLING and RELU_MAX_POOLING_BACKWARD is extremely useful for CP backend\n// by reducing unnecessary sparse-to-dense-to-sparse conversion.\n// For other backends, this operators is not necessary as it reduces an additional relu operator.\nif(OptimizerUtils.ALLOW_OPERATOR_FUSION && et == ExecType.CP && op == ConvOp.MAX_POOLING && isInputReLU(inputs.get(0))) {\n- in = inputs.get(0).getInput().get(0).constructLops();\n+ lhsInputLop = inputs.get(0).getInput().get(0).constructLops();\nlopOp = OperationTypes.RELU_MAX_POOLING;\n}\nelse if(OptimizerUtils.ALLOW_OPERATOR_FUSION && et == ExecType.CP && op == ConvOp.MAX_POOLING_BACKWARD && isInputReLU(inputs.get(0))) {\n- in = inputs.get(0).getInput().get(0).constructLops();\n+ lhsInputLop = inputs.get(0).getInput().get(0).constructLops();\nlopOp = OperationTypes.RELU_MAX_POOLING_BACKWARD;\n}\nelse if(OptimizerUtils.ALLOW_OPERATOR_FUSION && op == ConvOp.BIAS_ADD && isInputConv2d(inputs.get(0))) {\nlopOp = OperationTypes.DIRECT_CONV2D_BIAS_ADD;\n// the first lop is image\n- in = inputs.get(0).getInput().get(0).constructLops();\n+ lhsInputLop = inputs.get(0).getInput().get(0).constructLops();\n// the second lop is bias\n- in2 = inputs.get(1).constructLops();\n+ optionalRhsInputLop = inputs.get(1).constructLops();\n// Use the inputs from conv2d rather than bias_add\n- inputs1 = inputs.get(0).getInput();\n+ inputsOfPotentiallyFusedOp = inputs.get(0).getInput();\n}\nelse {\n- in = inputs.get(0).constructLops();\n+ lhsInputLop = inputs.get(0).constructLops();\n}\n+ // ---------------------------------------------------------------\n-// // TODO: Inserting reblock requires knowing columns apriori\n-// ConvolutionTransform transform1 = new ConvolutionTransform(addReblockIfNecessary(et, lopOp, in), lopOp, getDataType(), getValueType(), et, k);\n-// setReblockedOutputDimension(et, transform1);\n- double cpIntermediateMemEstimate = computeIntermediateMemEstimate(-1, -1, -1 );\n+ // ---------------------------------------------------------------\n+ // Compute intermediate memory budget that can be passed to GPU operators\n+ // for better CuDNN operator selection at runtime\n+ double intermediateMemEstimate = computeIntermediateMemEstimate(-1, -1, -1 );\nif(et == ExecType.GPU && _dim1 > 0 && _dim2 > 0) {\n// This enables us to compile more efficient matrix-matrix CuDNN operation instead of\n// row-by-row invocation of multiple vector-matrix CuDNN operations.\n// This is possible as the operations on GPU are single-threaded\ndouble optimisticIntermediateMemEstimate = GPUContextPool.initialGPUMemBudget() - getOutputMemEstimate() - inputs.get(0).getOutputMemEstimate();\n- if(in2 != null) {\n+ if(optionalRhsInputLop != null) {\noptimisticIntermediateMemEstimate -= inputs.get(1).getOutputMemEstimate();\n}\n- cpIntermediateMemEstimate = Math.max(cpIntermediateMemEstimate, optimisticIntermediateMemEstimate);\n+ intermediateMemEstimate = Math.max(intermediateMemEstimate, optimisticIntermediateMemEstimate);\n}\n- ConvolutionTransform transform1 = new ConvolutionTransform(in, lopOp, getDataType(), getValueType(), et, k, cpIntermediateMemEstimate);\n- setOutputDimensions(transform1);\n+ // ---------------------------------------------------------------\n- setLineNumbers(transform1);\n- in.addOutput(transform1);\n+ // Contruct the lop\n+ ConvolutionTransform convolutionLop = new ConvolutionTransform(lhsInputLop, lopOp,\n+ getDataType(), getValueType(), et, OptimizerUtils.getConstrainedNumThreads(_maxNumThreads), intermediateMemEstimate);\n- if(in2 != null) {\n- transform1.addInput(in2);\n- in2.addOutput(transform1);\n- }\n+ // Propagate the output dimensions and the line number of ConvolutionOp to ConvolutionTransform\n+ setOutputDimensions(convolutionLop);\n+ setLineNumbers(convolutionLop);\n- // stride1, stride2, padding1, padding2\n- // input_shape1, input_shape2, input_shape3, input_shape4,\n- // filter_shape1, filter_shape2, filter_shape3, filter_shape4\n- for( int i=1; i < inputs1.size(); i++ )\n- {\n- Lop ltmp = inputs1.get(i).constructLops();\n- transform1.addInput(ltmp);\n- ltmp.addOutput(transform1);\n- }\n- transform1.setLevel(); //force order of added lops\n- return transform1;\n+ // ---------------------------------------------------------------\n+ // Add input/output for parent lops of convolutionLop\n+ lhsInputLop.addOutput(convolutionLop);\n+ if(optionalRhsInputLop != null) {\n+ convolutionLop.addInput(optionalRhsInputLop);\n+ optionalRhsInputLop.addOutput(convolutionLop);\n+ }\n+ for( int i=1; i < inputsOfPotentiallyFusedOp.size(); i++ ) {\n+ Lop ltmp = inputsOfPotentiallyFusedOp.get(i).constructLops();\n+ convolutionLop.addInput(ltmp);\n+ ltmp.addOutput(convolutionLop);\n+ }\n+ // Only valid for MAX_POOLING_BACKWARD on GPU\n+ Lop optionalMaxPoolOutput = (et == ExecType.GPU) ? getMaxPoolOutputLop() : null;\n+ if(optionalMaxPoolOutput != null) {\n+ convolutionLop.addInput(optionalMaxPoolOutput);\n+ optionalMaxPoolOutput.addOutput(convolutionLop);\n+ }\n+ convolutionLop.setLevel(); //force order of added lops\n+ // ---------------------------------------------------------------\n+ return convolutionLop;\n}\n@@ -453,12 +513,10 @@ public class ConvolutionOp extends Hop implements MultiThreadedHop\nExecType REMOTE = OptimizerUtils.isSparkExecutionMode() ? ExecType.SPARK : ExecType.MR;\n- if( _etypeForced != null )\n- {\n+ if( _etypeForced != null ) {\n_etype = _etypeForced;\n}\n- else\n- {\n+ else {\nif ( OptimizerUtils.isMemoryBasedOptLevel() ) {\n_etype = findExecTypeByMemEstimate();\n}\n@@ -479,8 +537,9 @@ public class ConvolutionOp extends Hop implements MultiThreadedHop\nreturn _etype;\n}\n- // Caching parameters speed-ups dynamic recompilation time by avoiding unnecessary computeSizeInformation\n+ // Parameters recomputed in refreshSizeInformation and passed across many calls of getDim\nprivate ConvolutionParameters _cachedParams = new ConvolutionParameters(-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, _maxNumThreads);\n+\n// stride1, stride2, padding1, padding2\n// input_shape1, input_shape2, input_shape3, input_shape4,\n// filter_shape1, filter_shape2, filter_shape3, filter_shape4\n@@ -494,16 +553,16 @@ public class ConvolutionOp extends Hop implements MultiThreadedHop\nimageHeightHop = getInput().get(8);\nfilterHeightHop = getInput().get(12);\n_cachedParams.setIfUnknown(\n- getInput().get(6),\n- getInput().get(7),\n- imageHeightHop,\n- getInput().get(9),\n- getInput().get(10),\n- filterHeightHop,\n- getInput().get(13),\n- getInput().get(2),\n- getInput().get(3),\n- getInput().get(4),\n+ getInput().get(6), // N\n+ getInput().get(7), // C\n+ imageHeightHop, // H\n+ getInput().get(9), // W\n+ getInput().get(10), // K\n+ filterHeightHop, // R\n+ getInput().get(13), // S\n+ getInput().get(2), // stride_h\n+ getInput().get(3), // stride_w\n+ getInput().get(4), // pad+h\ngetInput().get(5), _maxNumThreads);\n}\nelse {\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/ConvolutionGPUInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/gpu/ConvolutionGPUInstruction.java", "diff": "@@ -92,8 +92,7 @@ public class ConvolutionGPUInstruction extends GPUInstruction {\nif( ( opcode.equalsIgnoreCase(\"conv2d\")\n|| opcode.equalsIgnoreCase(\"conv2d_backward_filter\")\n- || opcode.equalsIgnoreCase(\"conv2d_backward_data\")\n- || opcode.equalsIgnoreCase(\"maxpooling_backward\")) ) {\n+ || opcode.equalsIgnoreCase(\"conv2d_backward_data\")) ) {\nInstructionUtils.checkNumFields(parts, 16);\nCPOperand in1 = new CPOperand(parts[1]);\nCPOperand in2 = new CPOperand(parts[2]);\n@@ -119,6 +118,39 @@ public class ConvolutionGPUInstruction extends GPUInstruction {\nreturn new ConvolutionGPUInstruction(in1, in2, out, opcode, str, stride,\npadding, input_shape, filter_shape, Double.parseDouble(parts[16]));\n}\n+ else if( opcode.equalsIgnoreCase(\"maxpooling_backward\") ) {\n+ boolean withMaxPoolOut = false;\n+ if(parts.length == 18) {\n+ withMaxPoolOut = true;\n+ }\n+ else\n+ InstructionUtils.checkNumFields(parts, 16);\n+ CPOperand in1 = new CPOperand(parts[1]);\n+ CPOperand in2 = new CPOperand(parts[2]);\n+ CPOperand in3 = withMaxPoolOut ? new CPOperand(parts[15]) : null;\n+ CPOperand out = withMaxPoolOut ? new CPOperand(parts[16]) : new CPOperand(parts[15]);\n+ double memBudget = withMaxPoolOut ? Double.parseDouble(parts[17]) : Double.parseDouble(parts[16]);\n+\n+ ArrayList<CPOperand> stride = new ArrayList<>();\n+ ArrayList<CPOperand> padding = new ArrayList<>();\n+ ArrayList<CPOperand> input_shape = new ArrayList<>();\n+ ArrayList<CPOperand> filter_shape = new ArrayList<>();\n+ stride.add(new CPOperand(parts[3]));\n+ stride.add(new CPOperand(parts[4]));\n+ padding.add(new CPOperand(parts[5]));\n+ padding.add(new CPOperand(parts[6]));\n+ input_shape.add(new CPOperand(parts[7]));\n+ input_shape.add(new CPOperand(parts[8]));\n+ input_shape.add(new CPOperand(parts[9]));\n+ input_shape.add(new CPOperand(parts[10]));\n+ filter_shape.add(new CPOperand(parts[11]));\n+ filter_shape.add(new CPOperand(parts[12]));\n+ filter_shape.add(new CPOperand(parts[13]));\n+ filter_shape.add(new CPOperand(parts[14]));\n+\n+ return new ConvolutionGPUInstruction(in1, in2, in3, out, opcode, str, stride,\n+ padding, input_shape, filter_shape, memBudget);\n+ }\nelse if (opcode.equalsIgnoreCase(\"conv2d_bias_add\")) {\nInstructionUtils.checkNumFields(parts, 17);\nCPOperand in1 = new CPOperand(parts[1]);\n@@ -324,7 +356,7 @@ public class ConvolutionGPUInstruction extends GPUInstruction {\nelse if (instOpcode.equalsIgnoreCase(\"maxpooling_backward\")) {\nMatrixObject image = getMatrixInputForGPUInstruction(ec, _input1.getName());\nMatrixObject dout = getMatrixInputForGPUInstruction(ec, _input2.getName());\n-\n+ MatrixObject maxPoolOutput = _input3 != null ? getMatrixInputForGPUInstruction(ec, _input3.getName()) : null;\nif(dout.getNumRows() != N || dout.getNumColumns() != C*P*Q)\nthrow new DMLRuntimeException(\"Incorrect dimensions for dout in maxpooling_backward\");\nif(image.getNumRows() != N || image.getNumColumns() != C*H*W)\n@@ -333,7 +365,7 @@ public class ConvolutionGPUInstruction extends GPUInstruction {\nMatrixObject out = getDenseMatrixOutputForGPUInstruction(ec, _output.getName(), N, C * H * W);\n- LibMatrixCuDNN.maxpoolingBackward(ec.getGPUContext(0), getExtendedOpcode(), image, dout, out, N, C, H, W,\n+ LibMatrixCuDNN.maxpoolingBackward(ec.getGPUContext(0), getExtendedOpcode(), image, dout, maxPoolOutput, out, N, C, H, W,\nK, R, S, pad_h, pad_w, stride_h, stride_w, P, Q, _intermediateMemoryBudget);\n}\nelse {\n@@ -346,7 +378,8 @@ public class ConvolutionGPUInstruction extends GPUInstruction {\nif ( !instOpcode.equalsIgnoreCase(\"maxpooling\") )\nec.releaseMatrixInputForGPUInstruction(_input2.getName());\n- if (instOpcode.equalsIgnoreCase(\"conv2d_bias_add\"))\n+ if (instOpcode.equalsIgnoreCase(\"conv2d_bias_add\") ||\n+ (instOpcode.equalsIgnoreCase(\"maxpooling_backward\") && _input3 != null))\nec.releaseMatrixInputForGPUInstruction(_input3.getName());\nec.releaseMatrixOutputForGPUInstruction(_output.getName());\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixCuDNN.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixCuDNN.java", "diff": "@@ -519,6 +519,7 @@ public class LibMatrixCuDNN extends LibMatrixCUDA {\n* @param instName the invoking instruction's name for record {@link Statistics}.\n* @param image image as matrix object\n* @param dout delta matrix, output of previous layer\n+ * @param maxpoolOutput (optional and can be null) output of maxpool forward function\n* @param outputBlock output matrix\n* @param N batch size\n* @param C number of channels\n@@ -537,12 +538,14 @@ public class LibMatrixCuDNN extends LibMatrixCUDA {\n* @throws DMLRuntimeException if DMLRuntimeException occurs\n*/\npublic static void maxpoolingBackward(GPUContext gCtx, String instName, MatrixObject image, MatrixObject dout,\n- MatrixObject outputBlock, int N, int C, int H, int W, int K, int R,\n+ MatrixObject maxpoolOutput, MatrixObject outputBlock, int N, int C, int H, int W, int K, int R,\nint S, int pad_h, int pad_w, int stride_h, int stride_w, int P,\nint Q, double intermediateMemoryBudget) throws DMLRuntimeException {\nlong CHW = C*H*W; long CPQ = C*P*Q;\nlong NCHW = N*CHW; long NCPQ = N*CPQ;\n+ final boolean isMaxPoolOutputProvided = maxpoolOutput != null;\n+\nif(NCHW < maxNumElementsOfCuDNNTensor && NCPQ < maxNumElementsOfCuDNNTensor) {\n// Filter and output are accounted as dense in the memory estimation for conv2dBackwardData\nlong overhead = isInSparseFormat(gCtx, image) ? OptimizerUtils.estimateSizeExactSparsity(N, CHW, 1.0) : 0;\n@@ -551,19 +554,26 @@ public class LibMatrixCuDNN extends LibMatrixCUDA {\nif(overhead <= intermediateMemoryBudget) {\nPointer x = getDensePointerForCuDNN(gCtx, image, instName);\nPointer dy = getDensePointerForCuDNN(gCtx, dout, instName);\n- cudnnMaxpoolingBackward(gCtx, instName, x, dy, dx, N, C, H, W, K, R, S, pad_h, pad_w, stride_h, stride_w, P, Q);\n+ Pointer y = isMaxPoolOutputProvided ? getDensePointerForCuDNN(gCtx, maxpoolOutput, instName) : null;\n+ cudnnMaxpoolingBackward(gCtx, instName, x, dy, y, dx, N, C, H, W, K, R, S, pad_h, pad_w, stride_h, stride_w, P, Q);\n}\nelse {\nLibMatrixCuDNNInputRowFetcher imgFetcher = new LibMatrixCuDNNInputRowFetcher(gCtx, instName, image);\nLibMatrixCuDNNInputRowFetcher doutFetcher = new LibMatrixCuDNNInputRowFetcher(gCtx, instName, dout);\n+ LibMatrixCuDNNInputRowFetcher maxPoolOutFetcher = isMaxPoolOutputProvided ? new LibMatrixCuDNNInputRowFetcher(gCtx, instName, maxpoolOutput) : null;\nfor(int n = 0; n < N; n++) {\n- cudnnMaxpoolingBackward(gCtx, instName, imgFetcher.getNthRow(n), doutFetcher.getNthRow(n),\n+ Pointer x = imgFetcher.getNthRow(n);\n+ Pointer dy = doutFetcher.getNthRow(n);\n+ Pointer y = isMaxPoolOutputProvided ? maxPoolOutFetcher.getNthRow(n) : null;\n+ cudnnMaxpoolingBackward(gCtx, instName, x, dy, y,\ndx.withByteOffset(n*CHW*sizeOfDataType),\n1, C, H, W, K, R, S, pad_h, pad_w, stride_h, stride_w, P, Q);\n}\n// Deallocate temporary array to hold one element of input\nimgFetcher.close();\ndoutFetcher.close();\n+ if(isMaxPoolOutputProvided)\n+ maxPoolOutFetcher.close();\n}\n}\nelse {\n@@ -572,36 +582,33 @@ public class LibMatrixCuDNN extends LibMatrixCUDA {\n}\nprivate static void cudnnMaxpoolingBackward(GPUContext gCtx, String instName,\n- Pointer x, Pointer dy, Pointer dx,\n+ Pointer x, Pointer dy, Pointer y, Pointer dx,\nint N, int C, int H, int W, int K, int R,\nint S, int pad_h, int pad_w, int stride_h, int stride_w, int P,\nint Q) throws DMLRuntimeException {\nif(LOG.isTraceEnabled()) {\nLOG.trace(\"GPU : maxpoolingBackward\" + \", GPUContext=\" + gCtx);\n}\n- Pointer y = null;\n+\n+ boolean isMaxPoolOutputProvided = (y != null);\ntry(LibMatrixCuDNNPoolingDescriptors desc =\nLibMatrixCuDNNPoolingDescriptors.cudnnMaxpoolingBackwardDescriptors(gCtx, instName, N, C, H, W, K, R, S,\npad_h, pad_w, stride_h, stride_w, P, Q)) {\nlong t1=0, t2=0, t3=0;\n+ int status;\n+ if(!isMaxPoolOutputProvided) {\nif (GPUStatistics.DISPLAY_STATISTICS) t1 = System.nanoTime();\n-\n- // Calling PoolForward first, y is one of the inputs for poolBackward\n- // TODO: Remove calling poolForward after necessary changes at language level for poolBackward\nlong numBytes = N*C*P*Q*sizeOfDataType;\ny = gCtx.allocate(numBytes);\n-\nif (GPUStatistics.DISPLAY_STATISTICS) GPUStatistics.maintainCPMiscTimes(instName, GPUInstruction.MISC_TIMER_CUDNN_INIT, System.nanoTime() - t1);\n-\nif (GPUStatistics.DISPLAY_STATISTICS) t2 = System.nanoTime();\n- int status = cudnnPoolingForward(getCudnnHandle(gCtx), desc.poolingDesc, one(), desc.xDesc, x, zero(), desc.yDesc, y);\n+ status = cudnnPoolingForward(getCudnnHandle(gCtx), desc.poolingDesc, one(), desc.xDesc, x, zero(), desc.yDesc, y);\nif (GPUStatistics.DISPLAY_STATISTICS) GPUStatistics.maintainCPMiscTimes(instName, GPUInstruction.MISC_TIMER_MAXPOOLING_FORWARD_LIB, System.nanoTime() - t2);\n-\nif(status != jcuda.jcudnn.cudnnStatus.CUDNN_STATUS_SUCCESS) {\nthrow new DMLRuntimeException(\"Could not executed cudnnPoolingForward before cudnnPoolingBackward: \" + jcuda.jcudnn.cudnnStatus.stringFor(status));\n}\n-\n+ }\nif (GPUStatistics.DISPLAY_STATISTICS) t3 = System.nanoTime();\nstatus = cudnnPoolingBackward(getCudnnHandle(gCtx), desc.poolingDesc, one(), desc.yDesc, y, desc.dyDesc, dy, desc.xDesc, x, zero(), desc.dxDesc, dx);\nif (GPUStatistics.DISPLAY_STATISTICS) GPUStatistics.maintainCPMiscTimes(instName, GPUInstruction.MISC_TIMER_MAXPOOLING_BACKWARD_LIB, System.nanoTime() - t3);\n@@ -615,7 +622,7 @@ public class LibMatrixCuDNN extends LibMatrixCUDA {\nfinally {\nlong t4=0;\nif (GPUStatistics.DISPLAY_STATISTICS) t4 = System.nanoTime();\n- if(y != null)\n+ if(!isMaxPoolOutputProvided)\ngCtx.cudaFreeHelper(instName, y);\nif (GPUStatistics.DISPLAY_STATISTICS) GPUStatistics.maintainCPMiscTimes(instName, GPUInstruction.MISC_TIMER_CUDNN_CLEANUP, System.nanoTime() - t4);\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/gpu/NeuralNetworkOpTests.java", "new_path": "src/test/java/org/apache/sysml/test/gpu/NeuralNetworkOpTests.java", "diff": "@@ -575,6 +575,88 @@ public class NeuralNetworkOpTests extends GPUTests {\n+ }\n+ }\n+ }\n+ }\n+\n+\n+ @Test\n+ @Ignore\n+ public void testMaxPoolBackwardWithMaxpoolOut() {\n+ String scriptStr = \"tmp = max_pool(image, padding=[padH, padW], stride=[strideH, strideW], input_shape=[N,C,H,W], pool_size=[R,S]); print(sum(tmp)); O = max_pool_backward(image, dout, padding=[padH, padW], stride=[strideH, strideW], input_shape=[N,C,H,W], pool_size=[R,S])\";\n+\n+ for (long N : Nlst) {\n+ for (long C : Clst) {\n+ for (long H : Hlst) {\n+ long W = H;\n+ for (long R : Rlst) {\n+ long S = R;\n+ for (long strideH : strideLst) {\n+ long strideW = strideH;\n+ for (long padH : padLst) {\n+ long padW = padH;\n+ for (double sparsity : sparsitylst) {\n+\n+ // pool is smaller than image + padding\n+ if (R > (H + padH) || S > (W + padW))\n+ continue;\n+\n+ // Make sure ops fit in GPU memory and within constraints of cudnn\n+ long imageSize = N * C * H * W * 8l;\n+ if (imageSize > MAX_OP_SIZE) // image size\n+ continue;\n+ long poolSize = R * S * 8l;\n+ if (poolSize > MAX_OP_SIZE) // filter size\n+ continue;\n+\n+ int P = (int) ConvolutionUtils.getP(H, R, strideH, padH);\n+ int Q = (int) ConvolutionUtils.getQ(W, S, strideW, padW);\n+\n+ long doutSize = N * C * P * Q * 8l;\n+ if (doutSize > MAX_OP_SIZE) // dout/output size\n+ continue;\n+\n+ double imageSizeInMB = imageSize / (1024.0 * 1024.0);\n+ double poolSizeInMB = poolSize / (1024.0 * 1024.0);\n+ double doutSizeInMB = doutSize / (1024.0 * 1024.0);\n+ System.out\n+ .format(\"max_pool_backward, image[%d,%d,%d,%d](%.1fMB), pool[%d,%d](%.1f), dout[%d,%d,%d,%d](%.1fMB), stride[%d,%d], padding[%d,%d]\",\n+ N, C, H, W, imageSizeInMB, R, S, poolSizeInMB, N, C,\n+ P, Q, doutSizeInMB, strideH, strideW, padH, padW);\n+\n+ Matrix image = generateInputMatrix(spark, (int) N,\n+ (int) (C * H * W), -127.0, 127, sparsity, seed, true);\n+ Matrix dout = generateInputMatrix(spark, (int) N, (int) (C * P * Q),\n+ -127.0, 127, sparsity, seed, true);\n+ HashMap<String, Object> inputs = new HashMap<>();\n+ inputs.put(\"N\", N);\n+ inputs.put(\"C\", C);\n+ inputs.put(\"H\", H);\n+ inputs.put(\"W\", W);\n+ inputs.put(\"R\", R);\n+ inputs.put(\"S\", S);\n+ inputs.put(\"strideH\", strideH);\n+ inputs.put(\"strideW\", strideW);\n+ inputs.put(\"padH\", padH);\n+ inputs.put(\"padW\", padW);\n+ inputs.put(\"image\", image);\n+ inputs.put(\"dout\", dout);\n+ List<Object> outCPU = runOnCPU(spark, scriptStr, inputs,\n+ Arrays.asList(\"O\"));\n+ List<Object> outGPU = runOnGPU(spark, scriptStr, inputs,\n+ Arrays.asList(\"O\"));\n+ assertHeavyHitterPresent(\"gpu_maxpooling_backward\");\n+ assertEqualObjects(outCPU.get(0), outGPU.get(0));\n+ clearGPUMemory();\n+ }\n+ }\n+ }\n+ }\n+\n+\n+\n+\n}\n}\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofRowwise.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofRowwise.java", "diff": "@@ -39,7 +39,6 @@ import org.apache.sysml.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysml.runtime.matrix.data.SparseBlock;\nimport org.apache.sysml.runtime.matrix.data.SparseRow;\nimport org.apache.sysml.runtime.matrix.data.SparseRowVector;\n-import org.apache.sysml.runtime.util.UtilFunctions;\npublic abstract class SpoofRowwise extends SpoofOperator\n@@ -198,11 +197,9 @@ public abstract class SpoofRowwise extends SpoofOperator\n//core parallel execute\nExecutorService pool = Executors.newFixedThreadPool( k );\n- int nk = (a instanceof CompressedMatrixBlock) ? k :\n- UtilFunctions.roundToNext(Math.min(8*k,m/32), k);\n- int blklen = (int)(Math.ceil((double)m/nk));\n- if( a instanceof CompressedMatrixBlock )\n- blklen = BitmapEncoder.getAlignedBlocksize(blklen);\n+ ArrayList<Integer> blklens = (a instanceof CompressedMatrixBlock) ?\n+ LibMatrixMult.getAlignedBlockSizes(m, k, BitmapEncoder.BITMAP_BLOCK_SZ) :\n+ LibMatrixMult.getBalancedBlockSizesDefault(m, k, false);\ntry\n{\n@@ -210,8 +207,8 @@ public abstract class SpoofRowwise extends SpoofOperator\n//execute tasks\nArrayList<ParColAggTask> tasks = new ArrayList<>();\nint outLen = out.getNumRows() * out.getNumColumns();\n- for( int i=0; i<nk & i*blklen<m; i++ )\n- tasks.add(new ParColAggTask(a, b, scalars, n, n2, outLen, i*blklen, Math.min((i+1)*blklen, m)));\n+ for( int i=0, lb=0; i<blklens.size(); lb+=blklens.get(i), i++ )\n+ tasks.add(new ParColAggTask(a, b, scalars, n, n2, outLen, lb, lb+blklens.get(i)));\nList<Future<double[]>> taskret = pool.invokeAll(tasks);\n//aggregate partial results\nint len = _type.isColumnAgg() ? out.getNumRows()*out.getNumColumns() : 1;\n@@ -222,8 +219,8 @@ public abstract class SpoofRowwise extends SpoofOperator\nelse {\n//execute tasks\nArrayList<ParExecTask> tasks = new ArrayList<>();\n- for( int i=0; i<nk & i*blklen<m; i++ )\n- tasks.add(new ParExecTask(a, b, out, scalars, n, n2, i*blklen, Math.min((i+1)*blklen, m)));\n+ for( int i=0, lb=0; i<blklens.size(); lb+=blklens.get(i), i++ )\n+ tasks.add(new ParExecTask(a, b, out, scalars, n, n2, lb, lb+blklens.get(i)));\nList<Future<Long>> taskret = pool.invokeAll(tasks);\n//aggregate nnz, no need to aggregate results\nlong nnz = 0;\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixMult.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixMult.java", "diff": "@@ -209,8 +209,7 @@ public class LibMatrixMult\ntry {\nExecutorService pool = Executors.newFixedThreadPool( k );\nArrayList<MatrixMultTask> tasks = new ArrayList<>();\n- int nk = (pm2r||pm2c) ? k : UtilFunctions.roundToNext(Math.min(8*k,num/32), k);\n- ArrayList<Integer> blklens = getBalancedBlockSizes(num, nk);\n+ ArrayList<Integer> blklens = getBalancedBlockSizesDefault(num, k, (pm2r||pm2c));\nfor( int i=0, lb=0; i<blklens.size(); lb+=blklens.get(i), i++ )\ntasks.add(new MatrixMultTask(m1, m2, ret, tm2, pm2r, pm2c, lb, lb+blklens.get(i)));\n//execute tasks\n@@ -321,11 +320,10 @@ public class LibMatrixMult\n//(currently: always parallelization over number of rows)\ntry {\nExecutorService pool = Executors.newFixedThreadPool( k );\n+ ArrayList<Integer> blklens = getBalancedBlockSizesDefault(mX.rlen, k, true);\nArrayList<MatrixMultChainTask> tasks = new ArrayList<>();\n- int blklen = (int)(Math.ceil((double)mX.rlen/k));\n- blklen += (blklen%24 != 0)?24-blklen%24:0;\n- for( int i=0; i<k & i*blklen<mX.rlen; i++ )\n- tasks.add(new MatrixMultChainTask(mX, mV, mW, ct, i*blklen, Math.min((i+1)*blklen, mX.rlen)));\n+ for( int i=0, lb=0; i<blklens.size(); lb+=blklens.get(i), i++ )\n+ tasks.add(new MatrixMultChainTask(mX, mV, mW, ct, lb, lb+blklens.get(i)));\n//execute tasks\nList<Future<double[]>> taskret = pool.invokeAll(tasks);\npool.shutdown();\n@@ -1606,10 +1604,18 @@ public class LibMatrixMult\nfinal int blocksizeI = 24; // constraint: factor of 4\nfinal int blocksizeJ = 1024;\ndouble[] tmp = new double[blocksizeI];\n+ final int bn = (ru-rl) % blocksizeI;\n+\n+ //compute rest (not aligned to blocksize)\n+ for( int i=rl, aix=rl*cd; i < rl+bn; i++, aix+=cd ) {\n+ double val = dotProduct(a, b, aix, 0, cd);\n+ val *= (weights) ? w[i] : 1;\n+ val -= (weights2) ? w[i] : 0;\n+ vectMultiplyAdd(val, a, c, aix, 0, cd);\n+ }\n//blockwise mmchain computation\n- final int bn = ru - ru % blocksizeI; //rl blocksize aligned\n- for( int bi=rl; bi < bn; bi+=blocksizeI )\n+ for( int bi=rl+bn; bi < ru; bi+=blocksizeI )\n{\n//compute 1st matrix-vector for row block\nArrays.fill(tmp, 0);\n@@ -1633,14 +1639,6 @@ public class LibMatrixMult\na, c, aix, aix+cd, aix+2*cd, aix+3*cd, bj, bjmin);\n}\n}\n-\n- //compute rest (not aligned to blocksize)\n- for( int i=bn, aix=bn*cd; i < ru; i++, aix+=cd ) {\n- double val = dotProduct(a, b, aix, 0, cd);\n- val *= (weights) ? w[i] : 1;\n- val -= (weights2) ? w[i] : 0;\n- vectMultiplyAdd(val, a, c, aix, 0, cd);\n- }\n}\nprivate static void matrixMultChainSparse(MatrixBlock mX, MatrixBlock mV, MatrixBlock mW, MatrixBlock ret, ChainType ct, int rl, int ru)\n@@ -3578,7 +3576,7 @@ public class LibMatrixMult\npublic static boolean checkParColumnAgg(MatrixBlock m1, int k, boolean inclFLOPs) {\nreturn (8L * m1.clen * k <= MEM_OVERHEAD_THRESHOLD\n- && (!inclFLOPs || 4L * m1.rlen * m1.clen >= PAR_MINFLOP_THRESHOLD));\n+ && (!inclFLOPs || 4L * m1.rlen * m1.clen / (m1.sparse?2:1) >= PAR_MINFLOP_THRESHOLD));\n}\nprivate static boolean checkParMatrixMultRightInputRows( MatrixBlock m1, MatrixBlock m2, int k ) {\n@@ -3676,6 +3674,20 @@ public class LibMatrixMult\n}\n+ public static ArrayList<Integer> getBalancedBlockSizesDefault(int len, int k, boolean constK) {\n+ int nk = constK ? k : UtilFunctions.roundToNext(Math.min(8*k,len/32), k);\n+ return getBalancedBlockSizes(len, nk);\n+ }\n+\n+ public static ArrayList<Integer> getAlignedBlockSizes(int len, int k, int align) {\n+ int blklen = (int)(Math.ceil((double)len/k));\n+ blklen += ((blklen%align != 0) ? align-blklen%align : 0);\n+ ArrayList<Integer> ret = new ArrayList<>();\n+ for(int i=0; i<len; i+=blklen)\n+ ret.add(Math.min(blklen, len-i));\n+ return ret;\n+ }\n+\nprivate static ArrayList<Integer> getBalancedBlockSizes(int len, int k) {\nArrayList<Integer> ret = new ArrayList<>();\nint base = len / k;\n" } ]

[ { "change_type": "MODIFY", "old_path": "samples/jupyter-notebooks/DML Tips and Tricks (aka Fun With DML).ipynb", "new_path": "samples/jupyter-notebooks/DML Tips and Tricks (aka Fun With DML).ipynb", "diff": "\"cell_type\": \"markdown\",\n\"metadata\": {},\n\"source\": [\n- \"1. [Cross Validation](#CrossValidation)\\n\",\n+ \"1. [Replace NaN with mode](#NaN2Mode)\\n\",\n+ \"* [Use sample builtin function to create sample from matrix](#sample)\\n\",\n+ \"* [Count of Matching Values in two Matrices/Vectors](#MatchinRows)\\n\",\n+ \"* [Cross Validation](#CrossValidation)\\n\",\n\"* [Value-based join of two Matrices](#JoinMatrices)\\n\",\n\"* [Filter Matrix to include only Frequent Column Values](#FilterMatrix)\\n\",\n\"* [Construct (sparse) Matrix from (rowIndex, colIndex, values) triplets](#Construct_sparse_Matrix)\\n\",\n},\n{\n\"cell_type\": \"code\",\n- \"execution_count\": 2,\n+ \"execution_count\": 15,\n\"metadata\": {\n\"collapsed\": false,\n- \"scrolled\": true\n+ \"scrolled\": false\n},\n\"outputs\": [\n{\n\"name\": \"stdout\",\n\"output_type\": \"stream\",\n\"text\": [\n- \"2017-08-18 21:33:18 UTC\\n\"\n+ \"2017-09-22 07:57:57 UTC\\n\"\n]\n}\n],\n\"print (ml.buildTime())\"\n]\n},\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"## Replace NaN with mode<a id=\\\"NaN2Mode\\\" />\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"This functions replaces NaN in column with mode of column\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 13,\n+ \"metadata\": {\n+ \"collapsed\": false\n+ },\n+ \"outputs\": [\n+ {\n+ \"name\": \"stdout\",\n+ \"output_type\": \"stream\",\n+ \"text\": [\n+ \"Before: \\n\",\n+ \"1.000 NaN\\n\",\n+ \"1.000 NaN\\n\",\n+ \"1.000 2.000\\n\",\n+ \"2.000 1.000\\n\",\n+ \"1.000 2.000\\n\",\n+ \"\\n\",\n+ \"After: \\n\",\n+ \"1.000 2.000\\n\",\n+ \"1.000 2.000\\n\",\n+ \"1.000 2.000\\n\",\n+ \"2.000 1.000\\n\",\n+ \"1.000 2.000\\n\",\n+ \"\\n\",\n+ \"SystemML Statistics:\\n\",\n+ \"Total execution time:\\t\\t0.001 sec.\\n\",\n+ \"Number of executed Spark inst:\\t0.\\n\",\n+ \"\\n\",\n+ \"\\n\"\n+ ]\n+ }\n+ ],\n+ \"source\": [\n+ \"prog=\\\"\\\"\\\"\\n\",\n+ \"# Function for NaN-aware replacement with mode\\n\",\n+ \"replaceNaNwithMode = function (matrix[double] X, integer colId) \\n\",\n+ \" return (matrix[double] X) \\n\",\n+ \"{\\n\",\n+ \" Xi = replace (target=X[,colId], pattern=0/0, replacement=max(X[,colId])+1) # replace NaN with largest value + 1\\n\",\n+ \" agg = aggregate (target=Xi, groups=Xi, fn=\\\"count\\\") # count each distinct value\\n\",\n+ \" mode = as.scalar (rowIndexMax(t(agg[1:nrow(agg)-1, ]))) # mode is max frequent value except last value\\n\",\n+ \" X[,colId] = replace (target=Xi, pattern=max(Xi), replacement=mode) # fill in mode\\n\",\n+ \"}\\n\",\n+ \"\\n\",\n+ \"X = matrix('1 NaN 1 NaN 1 2 2 1 1 2', rows = 5, cols = 2)\\n\",\n+ \"\\n\",\n+ \"Y = replaceNaNwithMode (X, 2)\\n\",\n+ \"\\n\",\n+ \"print (\\\"Before: \\\\n\\\" + toString(X))\\n\",\n+ \"print (\\\"After: \\\\n\\\" + toString(Y))\\n\",\n+ \"\\\"\\\"\\\"\\n\",\n+ \"with jvm_stdout(True):\\n\",\n+ \" ml.execute(dml(prog))\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"## Use sample builtin function to create sample from matrix<a id=\\\"sample\\\" />\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"Use sample() function, create permutation matrix using table(), and pull sample from X.\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 18,\n+ \"metadata\": {\n+ \"collapsed\": false\n+ },\n+ \"outputs\": [\n+ {\n+ \"name\": \"stdout\",\n+ \"output_type\": \"stream\",\n+ \"text\": [\n+ \"X: \\n\",\n+ \"2.000 1.000\\n\",\n+ \"8.000 3.000\\n\",\n+ \"5.000 6.000\\n\",\n+ \"7.000 9.000\\n\",\n+ \"4.000 4.000\\n\",\n+ \"\\n\",\n+ \"sv: \\n\",\n+ \"1.000\\n\",\n+ \"4.000\\n\",\n+ \"\\n\",\n+ \"samples: \\n\",\n+ \"2.000 1.000\\n\",\n+ \"7.000 9.000\\n\",\n+ \"\\n\",\n+ \"SystemML Statistics:\\n\",\n+ \"Total execution time:\\t\\t0.001 sec.\\n\",\n+ \"Number of executed Spark inst:\\t0.\\n\",\n+ \"\\n\",\n+ \"\\n\"\n+ ]\n+ }\n+ ],\n+ \"source\": [\n+ \"prog=\\\"\\\"\\\"\\n\",\n+ \"X = matrix ('2 1 8 3 5 6 7 9 4 4', rows = 5, cols = 2 )\\n\",\n+ \"\\n\",\n+ \"nbrSamples = 2\\n\",\n+ \"\\n\",\n+ \"sv = order (target = sample (nrow (X), nbrSamples, FALSE)) # samples w/o replacement, and order \\n\",\n+ \"P = table (seq (1, nbrSamples), sv, nbrSamples, nrow(X)) # permutation matrix\\n\",\n+ \"samples = P %*% X; # apply P to perform selection\\n\",\n+ \"\\n\",\n+ \"\\n\",\n+ \"print (\\\"X: \\\\n\\\" + toString(X))\\n\",\n+ \"print (\\\"sv: \\\\n\\\" + toString(sv))\\n\",\n+ \"print (\\\"samples: \\\\n\\\" + toString(samples))\\n\",\n+ \"\\\"\\\"\\\"\\n\",\n+ \"with jvm_stdout(True):\\n\",\n+ \" ml.execute(dml(prog))\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"## Count of Matching Values in two Matrices/Vectors<a id=\\\"MatchingRows\\\" />\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"markdown\",\n+ \"metadata\": {},\n+ \"source\": [\n+ \"Given two matrices/vectors X and Y, get a count of the rows where X and Y have the same value.\"\n+ ]\n+ },\n+ {\n+ \"cell_type\": \"code\",\n+ \"execution_count\": 19,\n+ \"metadata\": {\n+ \"collapsed\": false\n+ },\n+ \"outputs\": [\n+ {\n+ \"name\": \"stdout\",\n+ \"output_type\": \"stream\",\n+ \"text\": [\n+ \"t(X): 8.000 4.000 5.000 4.000 9.000 10.000\\n\",\n+ \"\\n\",\n+ \"t(Y): 4.000 9.000 5.000 1.000 9.000 7.000\\n\",\n+ \"\\n\",\n+ \"Number of Matches: 2.0\\n\",\n+ \"\\n\",\n+ \"SystemML Statistics:\\n\",\n+ \"Total execution time:\\t\\t0.001 sec.\\n\",\n+ \"Number of executed Spark inst:\\t0.\\n\",\n+ \"\\n\",\n+ \"\\n\"\n+ ]\n+ }\n+ ],\n+ \"source\": [\n+ \"prog=\\\"\\\"\\\"\\n\",\n+ \"X = matrix('8 4 5 4 9 10', rows = 6, cols = 1)\\n\",\n+ \"Y = matrix('4 9 5 1 9 7 ', rows = 6, cols = 1)\\n\",\n+ \"\\n\",\n+ \"matches = sum (X == Y)\\n\",\n+ \"\\n\",\n+ \"print (\\\"t(X): \\\" + toString(t(X)))\\n\",\n+ \"print (\\\"t(Y): \\\" + toString(t(Y)))\\n\",\n+ \"print (\\\"Number of Matches: \\\" + matches + \\\"\\\\n\\\")\\n\",\n+ \"\\\"\\\"\\\"\\n\",\n+ \"with jvm_stdout(True):\\n\",\n+ \" ml.execute(dml(prog))\"\n+ ]\n+ },\n{\n\"cell_type\": \"markdown\",\n\"metadata\": {},\n\"cell_type\": \"code\",\n\"execution_count\": 4,\n\"metadata\": {\n- \"collapsed\": false\n+ \"collapsed\": true\n},\n\"outputs\": [\n{\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/LocalVariableMap.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/LocalVariableMap.java", "diff": "@@ -63,6 +63,10 @@ public class LocalVariableMap implements Cloneable\nreturn localMap.keySet();\n}\n+ public Set<Entry<String, Data>> entrySet() {\n+ return localMap.entrySet();\n+ }\n+\n/**\n* Retrieves the data object given its name.\n*\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/ParForProgramBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/ParForProgramBlock.java", "diff": "@@ -633,7 +633,7 @@ public class ParForProgramBlock extends ForProgramBlock\n//preserve shared input/result variables of cleanup\nArrayList<String> varList = ec.getVarList();\n- HashMap<String, Boolean> varState = ec.pinVariables(varList);\n+ boolean[] varState = ec.pinVariables(varList);\ntry\n{\n@@ -1329,7 +1329,7 @@ public class ParForProgramBlock extends ForProgramBlock\n}\n}\n- private void cleanupSharedVariables( ExecutionContext ec, HashMap<String,Boolean> varState )\n+ private void cleanupSharedVariables( ExecutionContext ec, boolean[] varState )\nthrows DMLRuntimeException\n{\n//TODO needs as precondition a systematic treatment of persistent read information.\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/ExecutionContext.java", "new_path": "src/main/java/org/apache/sysml/runtime/controlprogram/context/ExecutionContext.java", "diff": "package org.apache.sysml.runtime.controlprogram.context;\nimport java.util.ArrayList;\n-import java.util.HashMap;\nimport java.util.List;\nimport org.apache.commons.logging.Log;\n@@ -151,6 +150,11 @@ public class ExecutionContext {\nreturn _variables.get(name);\n}\n+ public Data getVariable(CPOperand operand) throws DMLRuntimeException {\n+ return operand.getDataType().isScalar() ?\n+ getScalarInput(operand) : getVariable(operand.getName());\n+ }\n+\npublic void setVariable(String name, Data val) {\n_variables.put(name, val);\n}\n@@ -528,30 +532,25 @@ public class ExecutionContext {\n* The function returns the OLD \"clean up\" state of matrix objects.\n*\n* @param varList variable list\n- * @return map of old cleanup state of matrix objects\n+ * @return indicator vector of old cleanup state of matrix objects\n*/\n- public HashMap<String,Boolean> pinVariables(ArrayList<String> varList)\n+ public boolean[] pinVariables(ArrayList<String> varList)\n{\n//2-pass approach since multiple vars might refer to same matrix object\n- HashMap<String, Boolean> varsState = new HashMap<>();\n+ boolean[] varsState = new boolean[varList.size()];\n//step 1) get current information\n- for( String var : varList )\n- {\n- Data dat = _variables.get(var);\n- if( dat instanceof MatrixObject ) {\n- MatrixObject mo = (MatrixObject)dat;\n- varsState.put( var, mo.isCleanupEnabled() );\n- }\n+ for( int i=0; i<varList.size(); i++ ) {\n+ Data dat = _variables.get(varList.get(i));\n+ if( dat instanceof MatrixObject )\n+ varsState[i] = ((MatrixObject)dat).isCleanupEnabled();\n}\n//step 2) pin variables\n- for( String var : varList ) {\n- Data dat = _variables.get(var);\n- if( dat instanceof MatrixObject ) {\n- MatrixObject mo = (MatrixObject)dat;\n- mo.enableCleanup(false);\n- }\n+ for( int i=0; i<varList.size(); i++ ) {\n+ Data dat = _variables.get(varList.get(i));\n+ if( dat instanceof MatrixObject )\n+ ((MatrixObject)dat).enableCleanup(false);\n}\nreturn varsState;\n@@ -573,11 +572,11 @@ public class ExecutionContext {\n* @param varList variable list\n* @param varsState variable state\n*/\n- public void unpinVariables(ArrayList<String> varList, HashMap<String,Boolean> varsState) {\n- for( String var : varList) {\n- Data dat = _variables.get(var);\n+ public void unpinVariables(ArrayList<String> varList, boolean[] varsState) {\n+ for( int i=0; i<varList.size(); i++ ) {\n+ Data dat = _variables.get(varList.get(i));\nif( dat instanceof MatrixObject )\n- ((MatrixObject)dat).enableCleanup(varsState.get(var));\n+ ((MatrixObject)dat).enableCleanup(varsState[i]);\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/FunctionCallCPInstruction.java", "new_path": "src/main/java/org/apache/sysml/runtime/instructions/cp/FunctionCallCPInstruction.java", "diff": "package org.apache.sysml.runtime.instructions.cp;\nimport java.util.ArrayList;\n-import java.util.Collection;\n-import java.util.HashMap;\nimport java.util.HashSet;\n-import java.util.LinkedList;\n+import java.util.Map.Entry;\nimport org.apache.sysml.api.DMLScript;\nimport org.apache.sysml.lops.Lop;\n@@ -43,6 +41,21 @@ import org.apache.sysml.runtime.instructions.InstructionUtils;\npublic class FunctionCallCPInstruction extends CPInstruction {\nprivate String _functionName;\nprivate String _namespace;\n+ private final CPOperand[] _boundInputs;\n+ private final ArrayList<String> _boundInputNames;\n+ private final ArrayList<String> _boundOutputNames;\n+ private HashSet<String> _expectRetVars = null;\n+\n+ private FunctionCallCPInstruction(String namespace, String functName, CPOperand[] boundInputs,\n+ ArrayList<String> boundInputNames, ArrayList<String> boundOutputNames, String istr) {\n+ super(null, functName, istr);\n+ _cptype = CPINSTRUCTION_TYPE.External;\n+ _functionName = functName;\n+ _namespace = namespace;\n+ _boundInputs = boundInputs;\n+ _boundInputNames = boundInputNames;\n+ _boundOutputNames = boundOutputNames;\n+ }\npublic String getFunctionName() {\nreturn _functionName;\n@@ -52,23 +65,6 @@ public class FunctionCallCPInstruction extends CPInstruction {\nreturn _namespace;\n}\n- // stores both the bound input and output parameters\n- private ArrayList<CPOperand> _boundInputParamOperands;\n- private ArrayList<String> _boundInputParamNames;\n- private ArrayList<String> _boundOutputParamNames;\n-\n- private FunctionCallCPInstruction(String namespace, String functName, ArrayList<CPOperand> boundInParamOperands,\n- ArrayList<String> boundInParamNames, ArrayList<String> boundOutParamNames, String istr) {\n- super(null, functName, istr);\n- _cptype = CPINSTRUCTION_TYPE.External;\n- _functionName = functName;\n- _namespace = namespace;\n- _boundInputParamOperands = boundInParamOperands;\n- _boundInputParamNames = boundInParamNames;\n- _boundOutputParamNames = boundOutParamNames;\n-\n- }\n-\npublic static FunctionCallCPInstruction parseInstruction(String str)\nthrows DMLRuntimeException\n{\n@@ -78,20 +74,17 @@ public class FunctionCallCPInstruction extends CPInstruction {\nString functionName = parts[2];\nint numInputs = Integer.valueOf(parts[3]);\nint numOutputs = Integer.valueOf(parts[4]);\n- ArrayList<CPOperand> boundInParamOperands = new ArrayList<>();\n- ArrayList<String> boundInParamNames = new ArrayList<>();\n- ArrayList<String> boundOutParamNames = new ArrayList<>();\n+ CPOperand[] boundInputs = new CPOperand[numInputs];\n+ ArrayList<String> boundInputNames = new ArrayList<>();\n+ ArrayList<String> boundOutputNames = new ArrayList<>();\nfor (int i = 0; i < numInputs; i++) {\n- CPOperand operand = new CPOperand(parts[5 + i]);\n- boundInParamOperands.add(operand);\n- boundInParamNames.add(operand.getName());\n+ boundInputs[i] = new CPOperand(parts[5 + i]);\n+ boundInputNames.add(boundInputs[i].getName());\n}\n- for (int i = 0; i < numOutputs; i++) {\n- boundOutParamNames.add(parts[5 + numInputs + i]);\n- }\n-\n- return new FunctionCallCPInstruction ( namespace,functionName,\n- boundInParamOperands, boundInParamNames, boundOutParamNames, str );\n+ for (int i = 0; i < numOutputs; i++)\n+ boundOutputNames.add(parts[5 + numInputs + i]);\n+ return new FunctionCallCPInstruction ( namespace,\n+ functionName, boundInputs, boundInputNames, boundOutputNames, str );\n}\n@Override\n@@ -120,10 +113,10 @@ public class FunctionCallCPInstruction extends CPInstruction {\n// get the function program block (stored in the Program object)\nFunctionProgramBlock fpb = ec.getProgram().getFunctionProgramBlock(_namespace, _functionName);\n- // sanity check number of function paramters\n- if( _boundInputParamNames.size() < fpb.getInputParams().size() ) {\n+ // sanity check number of function parameters\n+ if( _boundInputs.length < fpb.getInputParams().size() ) {\nthrow new DMLRuntimeException(\"Number of bound input parameters does not match the function signature \"\n- + \"(\"+_boundInputParamNames.size()+\", but \"+fpb.getInputParams().size()+\" expected)\");\n+ + \"(\"+_boundInputs.length+\", but \"+fpb.getInputParams().size()+\" expected)\");\n}\n// create bindings to formal parameters for given function call\n@@ -131,35 +124,31 @@ public class FunctionCallCPInstruction extends CPInstruction {\nLocalVariableMap functionVariables = new LocalVariableMap();\nfor( int i=0; i<fpb.getInputParams().size(); i++)\n{\n- DataIdentifier currFormalParam = fpb.getInputParams().get(i);\n- String currFormalParamName = currFormalParam.getName();\n- Data currFormalParamValue = null;\n-\n- CPOperand operand = _boundInputParamOperands.get(i);\n- String varname = operand.getName();\n//error handling non-existing variables\n- if( !operand.isLiteral() && !ec.containsVariable(varname) ) {\n- throw new DMLRuntimeException(\"Input variable '\"+varname+\"' not existing on call of \" +\n+ CPOperand input = _boundInputs[i];\n+ if( !input.isLiteral() && !ec.containsVariable(input.getName()) ) {\n+ throw new DMLRuntimeException(\"Input variable '\"+input.getName()+\"' not existing on call of \" +\nDMLProgram.constructFunctionKey(_namespace, _functionName) + \" (line \"+getLineNum()+\").\");\n}\n//get input matrix/frame/scalar\n- currFormalParamValue = (operand.getDataType()!=DataType.SCALAR) ? ec.getVariable(varname) :\n- ec.getScalarInput(varname, operand.getValueType(), operand.isLiteral());\n+ DataIdentifier currFormalParam = fpb.getInputParams().get(i);\n+ Data value = ec.getVariable(input);\n//graceful value type conversion for scalar inputs with wrong type\n- if( currFormalParamValue.getDataType() == DataType.SCALAR\n- && currFormalParamValue.getValueType() != currFormalParam.getValueType() )\n+ if( value.getDataType() == DataType.SCALAR\n+ && value.getValueType() != currFormalParam.getValueType() )\n{\n- currFormalParamValue = ScalarObjectFactory.createScalarObject(\n- currFormalParam.getValueType(), (ScalarObject) currFormalParamValue);\n+ value = ScalarObjectFactory.createScalarObject(\n+ currFormalParam.getValueType(), (ScalarObject)value);\n}\n- functionVariables.put(currFormalParamName, currFormalParamValue);\n+ //set input parameter\n+ functionVariables.put(currFormalParam.getName(), value);\n}\n// Pin the input variables so that they do not get deleted\n// from pb's symbol table at the end of execution of function\n- HashMap<String,Boolean> pinStatus = ec.pinVariables(_boundInputParamNames);\n+ boolean[] pinStatus = ec.pinVariables(_boundInputNames);\n// Create a symbol table under a new execution context for the function invocation,\n// and copy the function arguments into the created table.\n@@ -182,29 +171,29 @@ public class FunctionCallCPInstruction extends CPInstruction {\nString fname = DMLProgram.constructFunctionKey(_namespace, _functionName);\nthrow new DMLRuntimeException(\"error executing function \" + fname, e);\n}\n- LocalVariableMap retVars = fn_ec.getVariables();\n// cleanup all returned variables w/o binding\n- Collection<String> retVarnames = new LinkedList<>(retVars.keySet());\n- HashSet<String> probeVars = new HashSet<>();\n+ if( _expectRetVars == null ) {\n+ _expectRetVars = new HashSet<>();\nfor(DataIdentifier di : fpb.getOutputParams())\n- probeVars.add(di.getName());\n- for( String var : retVarnames ) {\n- if( !probeVars.contains(var) ) //cleanup candidate\n- {\n- Data dat = fn_ec.removeVariable(var);\n- if( dat != null && dat instanceof MatrixObject )\n- fn_ec.cleanupMatrixObject((MatrixObject)dat);\n+ _expectRetVars.add(di.getName());\n}\n+\n+ LocalVariableMap retVars = fn_ec.getVariables();\n+ for( Entry<String,Data> var : retVars.entrySet() ) {\n+ if( _expectRetVars.contains(var.getKey()) )\n+ continue;\n+ //cleanup unexpected return values to avoid leaks\n+ if( var.getValue() instanceof MatrixObject )\n+ fn_ec.cleanupMatrixObject((MatrixObject)var.getValue());\n}\n// Unpin the pinned variables\n- ec.unpinVariables(_boundInputParamNames, pinStatus);\n+ ec.unpinVariables(_boundInputNames, pinStatus);\n// add the updated binding for each return variable to the variables in original symbol table\nfor (int i=0; i< fpb.getOutputParams().size(); i++){\n-\n- String boundVarName = _boundOutputParamNames.get(i);\n+ String boundVarName = _boundOutputNames.get(i);\nData boundValue = retVars.get(fpb.getOutputParams().get(i).getName());\nif (boundValue == null)\nthrow new DMLRuntimeException(boundVarName + \" was not assigned a return value\");\n@@ -240,14 +229,12 @@ public class FunctionCallCPInstruction extends CPInstruction {\nLOG.debug(\"ExternalBuiltInFunction: \" + this.toString());\n}\n- public ArrayList<String> getBoundInputParamNames()\n- {\n- return _boundInputParamNames;\n+ public ArrayList<String> getBoundInputParamNames() {\n+ return _boundInputNames;\n}\n- public ArrayList<String> getBoundOutputParamNames()\n- {\n- return _boundOutputParamNames;\n+ public ArrayList<String> getBoundOutputParamNames() {\n+ return _boundOutputNames;\n}\npublic void setFunctionName(String fname)\n@@ -277,6 +264,4 @@ public class FunctionCallCPInstruction extends CPInstruction {\nreturn sb.substring( 0, sb.length()-Lop.OPERAND_DELIMITOR.length() );\n}\n-\n-\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/compress/CompressedLinregCG.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/compress/CompressedLinregCG.java", "diff": "@@ -100,23 +100,22 @@ public class CompressedLinregCG extends AutomatedTestBase\ntry\n{\n- String TEST_NAME = testname;\n- TestConfiguration config = getTestConfiguration(TEST_NAME);\n+ TestConfiguration config = getTestConfiguration(testname);\n+ loadTestConfiguration(config);\n/* This is for running the junit test the new way, i.e., construct the arguments directly */\n- String HOME = SCRIPT_DIR + \"functions/codegen/\";\n+ String HOME1 = SCRIPT_DIR + \"functions/compress/\";\n+ String HOME2 = SCRIPT_DIR + \"functions/codegen/\";\nfullDMLScriptName = \"scripts/algorithms/LinearRegCG.dml\";\nprogramArgs = new String[]{ \"-explain\", \"-stats\", \"-nvargs\", \"X=\"+input(\"X\"), \"Y=\"+input(\"y\"),\n\"icpt=\"+String.valueOf(intercept), \"tol=\"+String.valueOf(epsilon),\n\"maxi=\"+String.valueOf(maxiter), \"reg=\"+String.valueOf(regular), \"B=\"+output(\"w\")};\n- fullRScriptName = HOME + \"Algorithm_LinregCG.R\";\n+ fullRScriptName = HOME2 + \"Algorithm_LinregCG.R\";\nrCmd = \"Rscript\" + \" \" + fullRScriptName + \" \" +\n- HOME + INPUT_DIR + \" \" +\n+ HOME1 + INPUT_DIR + \" \" +\nString.valueOf(intercept) + \" \" + String.valueOf(epsilon) + \" \" +\n- String.valueOf(maxiter) + \" \" + String.valueOf(regular) + HOME + EXPECTED_DIR;\n-\n- loadTestConfiguration(config);\n+ String.valueOf(maxiter) + \" \" + String.valueOf(regular) + \" \"+ HOME1 + EXPECTED_DIR;\n//generate actual datasets\ndouble[][] X = getRandomMatrix(rows, cols, 1, 1, sparse?sparsity2:sparsity1, 7);\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/CodegenUtils.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/CodegenUtils.java", "diff": "@@ -46,7 +46,10 @@ import org.apache.sysml.api.DMLScript;\nimport org.apache.sysml.hops.codegen.SpoofCompiler;\nimport org.apache.sysml.hops.codegen.SpoofCompiler.CompilerType;\nimport org.apache.sysml.runtime.DMLRuntimeException;\n+import org.apache.sysml.runtime.codegen.SpoofOperator.SideInput;\n+import org.apache.sysml.runtime.codegen.SpoofOperator.SideInputSparseCell;\nimport org.apache.sysml.runtime.io.IOUtilFunctions;\n+import org.apache.sysml.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysml.runtime.util.LocalFileUtils;\nimport org.apache.sysml.utils.Statistics;\nimport org.codehaus.janino.SimpleCompiler;\n@@ -152,6 +155,13 @@ public class CodegenUtils\nreturn ret;\n}\n+ public static SideInput createSideInput(MatrixBlock in) {\n+ SideInput ret = (in.isInSparseFormat() || !in.isAllocated()) ?\n+ new SideInput(null, in, in.getNumColumns()) :\n+ new SideInput(in.getDenseBlock(), null, in.getNumColumns());\n+ return (ret.mdat != null) ? new SideInputSparseCell(ret) : ret;\n+ }\n+\n////////////////////////////\n//JANINO-specific methods (used for spark environments)\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofOperator.java", "new_path": "src/main/java/org/apache/sysml/runtime/codegen/SpoofOperator.java", "diff": "@@ -249,6 +249,10 @@ public abstract class SpoofOperator implements Serializable\npublic double[] values(int r) {\nreturn ddat;\n}\n+ public double getValue(int r, int c) {\n+ return SpoofOperator.getValue(this, clen, r, c);\n+ }\n+ public void reset() {}\n}\npublic static class SideInputSparseRow extends SideInput {\n@@ -315,5 +319,9 @@ public abstract class SpoofOperator implements Serializable\nreturn (currColPos < currLen && indexes[currColPos]==colIndex) ?\nvalues[currColPos] : 0;\n}\n+ @Override\n+ public void reset() {\n+ currColPos = 0;\n+ }\n}\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/udf/lib/RowClassMeet.java", "new_path": "src/main/java/org/apache/sysml/udf/lib/RowClassMeet.java", "diff": "package org.apache.sysml.udf.lib;\nimport java.io.IOException;\n-import java.util.ArrayList;\n-import java.util.Arrays;\n-import java.util.Comparator;\n-import java.util.Iterator;\n+import java.util.HashMap;\nimport java.util.Map.Entry;\n-import java.util.TreeMap;\nimport org.apache.sysml.runtime.DMLRuntimeException;\n-import org.apache.sysml.runtime.controlprogram.caching.CacheException;\n-import org.apache.sysml.runtime.matrix.data.IJV;\n+import org.apache.sysml.runtime.codegen.CodegenUtils;\n+import org.apache.sysml.runtime.codegen.SpoofOperator.SideInput;\n+import org.apache.sysml.runtime.compress.utils.IntArrayList;\nimport org.apache.sysml.runtime.matrix.data.InputInfo;\nimport org.apache.sysml.runtime.matrix.data.MatrixBlock;\nimport org.apache.sysml.runtime.matrix.data.OutputInfo;\n+import org.apache.sysml.runtime.util.UtilFunctions;\nimport org.apache.sysml.udf.FunctionParameter;\nimport org.apache.sysml.udf.Matrix;\nimport org.apache.sysml.udf.PackageFunction;\nimport org.apache.sysml.udf.Matrix.ValueType;\n+\n/**\n* Performs following operation:\n* # Computes the intersection (\"meet\") of equivalence classes for\n@@ -73,8 +72,6 @@ public class RowClassMeet extends PackageFunction {\nprivate static final long serialVersionUID = 1L;\nprivate Matrix CMat, NMat;\n- private MatrixBlock A, B, C, N;\n- private int nr, nc;\n@Override\npublic int getNumFunctionOutputs() {\n@@ -83,124 +80,75 @@ public class RowClassMeet extends PackageFunction {\n@Override\npublic FunctionParameter getFunctionOutput(int pos) {\n- if(pos == 0)\n- return CMat;\n- else if(pos == 1)\n- return NMat;\n- else\n+ switch( pos ) {\n+ case 0: return CMat;\n+ case 1: return NMat;\n+ default:\nthrow new RuntimeException(\"RowClassMeet produces only one output\");\n}\n-\n-\n- public class ClassLabels {\n- public double aVal;\n- public double bVal;\n- public ClassLabels(double aVal, double bVal) {\n- this.aVal = aVal;\n- this.bVal = bVal;\n- }\n- }\n-\n- public class ClassLabelComparator implements Comparator<ClassLabels> {\n- Integer tmp1, tmp2;\n- @Override\n- public int compare(ClassLabels o1, ClassLabels o2) {\n- if(o1.aVal != o2.aVal) {\n- tmp1 = (int) o1.aVal;\n- tmp2 = (int) o2.aVal;\n- }\n- else {\n- tmp1 = (int) o1.bVal;\n- tmp2 = (int) o2.bVal;\n- }\n- return tmp1.compareTo(tmp2);\n- }\n- }\n-\n- double [] getRow(MatrixBlock B, double [] bRow, int i) {\n- if(B.getNumRows() == 1)\n- i = 0;\n- Arrays.fill(bRow, 0);\n- if(B.isInSparseFormat()) {\n- Iterator<IJV> iter = B.getSparseBlockIterator(i, i+1);\n- while(iter.hasNext()) {\n- IJV ijv = iter.next();\n- bRow[ijv.getJ()] = ijv.getV();\n- }\n- }\n- else {\n- double [] denseBlk = B.getDenseBlock();\n- if(denseBlk != null)\n- System.arraycopy(denseBlk, i*B.getNumColumns(), bRow, 0, B.getNumColumns());\n- }\n- return bRow;\n}\n@Override\npublic void execute() {\n- try {\n- A = ((Matrix) getFunctionInput(0)).getMatrixObject().acquireRead();\n- B = ((Matrix) getFunctionInput(1)).getMatrixObject().acquireRead();\n- nr = Math.max(A.getNumRows(), B.getNumRows());\n- nc = Math.max(A.getNumColumns(), B.getNumColumns());\n-\n- double [] bRow = new double[B.getNumColumns()];\n- CMat = new Matrix( createOutputFilePathAndName( \"TMP\" ), nr, nc, ValueType.Double );\n- C = new MatrixBlock(nr, nc, false);\n- C.allocateDenseBlock();\n- NMat = new Matrix( createOutputFilePathAndName( \"TMP\" ), nr, nc, ValueType.Double );\n- N = new MatrixBlock(nr, nc, false);\n- N.allocateDenseBlock();\n-\n- double [] cBlk = C.getDenseBlock();\n- double [] nBlk = N.getDenseBlock();\n-\n- if(B.getNumRows() == 1)\n- getRow(B, bRow, 0);\n-\n- for(int i = 0; i < A.getNumRows(); i++) {\n- if(B.getNumRows() != 1)\n- getRow(B, bRow, i);\n-\n- // Create class labels\n- TreeMap<ClassLabels, ArrayList<Integer>> classLabelMapping = new TreeMap<>(new ClassLabelComparator());\n+ try\n+ {\n+ MatrixBlock A = ((Matrix) getFunctionInput(0)).getMatrixObject().acquireRead();\n+ MatrixBlock B = ((Matrix) getFunctionInput(1)).getMatrixObject().acquireRead();\n+ int nr = Math.max(A.getNumRows(), B.getNumRows());\n+ int nc = Math.max(A.getNumColumns(), B.getNumColumns());\n+ MatrixBlock C = new MatrixBlock(nr, nc, false).allocateBlock();\n+ MatrixBlock N = new MatrixBlock(nr, nc, false).allocateBlock();\n+ double[] dC = C.getDenseBlock();\n+ double[] dN = N.getDenseBlock();\n+ //wrap both A and B into side inputs for efficient sparse access\n+ SideInput sB = CodegenUtils.createSideInput(B);\n+ boolean mv = (B.getNumRows() == 1);\n+ int numCols = Math.min(A.getNumColumns(), B.getNumColumns());\n+\n+ HashMap<ClassLabel, IntArrayList> classLabelMapping = new HashMap<>();\n+\n+ for(int i=0, ai=0; i < A.getNumRows(); i++, ai+=A.getNumColumns()) {\n+ classLabelMapping.clear(); sB.reset();\nif( A.isInSparseFormat() ) {\n- Iterator<IJV> iter = A.getSparseBlockIterator(i, i+1);\n- while(iter.hasNext()) {\n- IJV ijv = iter.next();\n- int j = ijv.getJ();\n- double aVal = ijv.getV();\n- if(aVal != 0 && bRow[j] != 0) {\n- ClassLabels key = new ClassLabels(aVal, bRow[j]);\n+ if(A.getSparseBlock()==null || A.getSparseBlock().isEmpty(i))\n+ continue;\n+ int alen = A.getSparseBlock().size(i);\n+ int apos = A.getSparseBlock().pos(i);\n+ int[] aix = A.getSparseBlock().indexes(i);\n+ double[] avals = A.getSparseBlock().values(i);\n+ for(int k=apos; k<apos+alen; k++) {\n+ if( aix[k] >= numCols ) break;\n+ int bval = (int)sB.getValue(mv?0:i, aix[k]);\n+ if( bval != 0 ) {\n+ ClassLabel key = new ClassLabel((int)avals[k], bval);\nif(!classLabelMapping.containsKey(key))\n- classLabelMapping.put(key, new ArrayList<Integer>());\n- classLabelMapping.get(key).add(j);\n+ classLabelMapping.put(key, new IntArrayList());\n+ classLabelMapping.get(key).appendValue(aix[k]);\n}\n}\n}\nelse {\ndouble [] denseBlk = A.getDenseBlock();\n- if(denseBlk != null) {\n- int offset = i*A.getNumColumns();\n- for(int j = 0; j < A.getNumColumns(); j++) {\n- double aVal = denseBlk[offset + j];\n- if(aVal != 0 && bRow[j] != 0) {\n- ClassLabels key = new ClassLabels(aVal, bRow[j]);\n+ if(denseBlk == null) break;\n+ for(int j = 0; j < numCols; j++) {\n+ int aVal = (int) denseBlk[ai+j];\n+ int bVal = (int) sB.getValue(mv?0:i, j);\n+ if(aVal != 0 && bVal != 0) {\n+ ClassLabel key = new ClassLabel(aVal, bVal);\nif(!classLabelMapping.containsKey(key))\n- classLabelMapping.put(key, new ArrayList<Integer>());\n- classLabelMapping.get(key).add(j);\n- }\n+ classLabelMapping.put(key, new IntArrayList());\n+ classLabelMapping.get(key).appendValue(j);\n}\n}\n}\n-\nint labelID = 1;\n- for(Entry<ClassLabels, ArrayList<Integer>> entry : classLabelMapping.entrySet()) {\n- double nVal = entry.getValue().size();\n- for(Integer j : entry.getValue()) {\n- nBlk[i*nc + j] = nVal;\n- cBlk[i*nc + j] = labelID;\n+ for(Entry<ClassLabel, IntArrayList> entry : classLabelMapping.entrySet()) {\n+ int nVal = entry.getValue().size();\n+ int[] list = entry.getValue().extractValues();\n+ for(int k=0, off=i*nc; k<nVal; k++) {\n+ dN[off+list[k]] = nVal;\n+ dC[off+list[k]] = labelID;\n}\nlabelID++;\n}\n@@ -208,23 +156,37 @@ public class RowClassMeet extends PackageFunction {\n((Matrix) getFunctionInput(0)).getMatrixObject().release();\n((Matrix) getFunctionInput(1)).getMatrixObject().release();\n- } catch (CacheException e) {\n- throw new RuntimeException(\"Error while executing RowClassMeet\", e);\n- }\n- try {\n- C.recomputeNonZeros();\n- C.examSparsity();\n+ //prepare outputs\n+ C.recomputeNonZeros(); C.examSparsity();\n+ CMat = new Matrix( createOutputFilePathAndName( \"TMP\" ), nr, nc, ValueType.Double );\nCMat.setMatrixDoubleArray(C, OutputInfo.BinaryBlockOutputInfo, InputInfo.BinaryBlockInputInfo);\n- N.recomputeNonZeros();\n- N.examSparsity();\n+ N.recomputeNonZeros(); N.examSparsity();\n+ NMat = new Matrix( createOutputFilePathAndName( \"TMP\" ), nr, nc, ValueType.Double );\nNMat.setMatrixDoubleArray(N, OutputInfo.BinaryBlockOutputInfo, InputInfo.BinaryBlockInputInfo);\n- } catch (DMLRuntimeException e) {\n- throw new RuntimeException(\"Error while executing RowClassMeet\", e);\n- } catch (IOException e) {\n+ }\n+ catch (DMLRuntimeException | IOException e) {\nthrow new RuntimeException(\"Error while executing RowClassMeet\", e);\n}\n}\n-\n+ private static class ClassLabel {\n+ public int aVal;\n+ public int bVal;\n+ public ClassLabel(int aVal, int bVal) {\n+ this.aVal = aVal;\n+ this.bVal = bVal;\n+ }\n+ @Override\n+ public int hashCode() {\n+ return UtilFunctions.intHashCode(aVal, bVal);\n+ }\n+ @Override\n+ public boolean equals(Object o) {\n+ if( !(o instanceof ClassLabel) )\n+ return false;\n+ ClassLabel that = (ClassLabel) o;\n+ return aVal == that.aVal && bVal == that.bVal;\n+ }\n+ }\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "pom.xml", "new_path": "pom.xml", "diff": "<?xml version=\"1.0\" encoding=\"UTF-8\"?>\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-<project xmlns=\"http://maven.apache.org/POM/4.0.0\" xmlns:xsi=\"http://www.w3.org/2001/XMLSchema-instance\" xsi:schemaLocation=\"http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd\">\n+<!-- * Licensed to the Apache Software Foundation (ASF) under one * or more\n+ contributor license agreements. See the NOTICE file * distributed with this\n+ work for additional information * regarding copyright ownership. The ASF\n+ licenses this file * to you under the Apache License, Version 2.0 (the *\n+ \"License\"); you may not use this file except in compliance * with the License.\n+ You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0\n+ * * Unless required by applicable law or agreed to in writing, * software\n+ distributed under the License is distributed on an * \"AS IS\" BASIS, WITHOUT\n+ WARRANTIES OR CONDITIONS OF ANY * KIND, either express or implied. See the\n+ License for the * specific language governing permissions and limitations\n+ * under the License. -->\n+<project xmlns=\"http://maven.apache.org/POM/4.0.0\" xmlns:xsi=\"http://www.w3.org/2001/XMLSchema-instance\"\n+ xsi:schemaLocation=\"http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd\">\n<modelVersion>4.0.0</modelVersion>\n<parent>\n<groupId>org.apache</groupId>\n<build>\n- <!-- Adds scripts to main jar, in-memory jar, sources jar, and standalone jar -->\n+ <!-- Adds scripts to main jar, in-memory jar, sources jar, and standalone\n+ jar -->\n<resources>\n<resource>\n<directory>scripts</directory>\n<executions>\n<execution>\n<id>default-jar</id>\n- <goals><goal>jar</goal></goals>\n+ <goals>\n+ <goal>jar</goal>\n+ </goals>\n<phase>package</phase>\n<configuration>\n<archive>\n<!-- Third argument prevents Java from popping up lots of windows on\nMacOS -->\n- <argLine>-Dfile.encoding=UTF-8 -Xmx2g -Xms2g -Xmn200m -Djava.awt.headless=true</argLine>\n+ <argLine>-Dfile.encoding=UTF-8 -Xmx2g -Xms2g -Xmn200m\n+ -Djava.awt.headless=true</argLine>\n<includes>\n<!-- All tests are integration tests as far as Maven is concerned. -->\n</configuration>\n</execution>\n<execution>\n- <!-- NOTE: We don't clean up systemml.egg-info since this makes it difficult\n- to uninstall a dev version after a maven clean -->\n+ <!-- NOTE: We don't clean up systemml.egg-info since this makes it\n+ difficult to uninstall a dev version after a maven clean -->\n<id>clean-python-files</id>\n<phase>clean</phase>\n<goals>\n<phase>package</phase>\n<configuration>\n<target name=\"copy and rename JAR\">\n- <copy file=\"${project.build.directory}/${project.artifactId}-${project.version}.jar\" tofile=\"${project.build.directory}/SystemML.jar\" />\n+ <copy\n+ file=\"${project.build.directory}/${project.artifactId}-${project.version}.jar\"\n+ tofile=\"${project.build.directory}/SystemML.jar\" />\n</target>\n</configuration>\n<goals>\n<profile>\n<!-- Can be used to ignore doclint javadoc issues -->\n<id>ignore-doclint</id>\n- <!-- <activation>\n- <jdk>[1.8,)</jdk>\n- </activation> -->\n+ <!-- <activation> <jdk>[1.8,)</jdk> </activation> -->\n<properties>\n<javadoc.opts>-Xdoclint:none</javadoc.opts>\n</properties>\n</profile>\n<profile>\n- <!-- Profile to create binary distributions.\n- Execute with `mvn clean package -P distribution` -->\n+ <!-- Profile to create binary distributions. Execute with `mvn clean package\n+ -P distribution` -->\n<id>distribution</id>\n<build>\n<plugins>\n<artifactId>maven-javadoc-plugin</artifactId>\n<version>2.10.3</version>\n<configuration>\n- <!-- Need to include the following packages, so exclude others:\n- org.apache.sysml.api\n- org.apache.sysml.runtime.instructions.spark.utils (for RDDConverterUtils, etc)\n- org.apache.sysml.runtime.matrix (for MatrixCharacteristics, etc)\n- org.apache.sysml.runtime.matrix.data (for MatrixIndexes, MatrixBlock, etc)\n- org.apache.sysml.udf\n- -->\n+ <!-- Need to include the following packages, so exclude others: org.apache.sysml.api\n+ org.apache.sysml.runtime.instructions.spark.utils (for RDDConverterUtils,\n+ etc) org.apache.sysml.runtime.matrix (for MatrixCharacteristics, etc) org.apache.sysml.runtime.matrix.data\n+ (for MatrixIndexes, MatrixBlock, etc) org.apache.sysml.udf -->\n<excludePackageNames>caffe:org.apache.sysml.conf:org.apache.sysml.debug:org.apache.sysml.hops:org.apache.sysml.lops:org.apache.sysml.parser:org.apache.sysml.runtime.controlprogram:org.apache.sysml.runtime.functionobjects:org.apache.sysml.runtime.instructions.cp:org.apache.sysml.runtime.instructions.cpfile:org.apache.sysml.runtime.instructions.mr:org.apache.sysml.runtime.instructions.spark.data:org.apache.sysml.runtime.instructions.spark.functions:org.apache.sysml.runtime.io:org.apache.sysml.runtime.matrix.data.hadoopfix:org.apache.sysml.runtime.matrix.mapred:org.apache.sysml.runtime.matrix.operators:org.apache.sysml.runtime.matrix.sort:org.apache.sysml.runtime.transform:org.apache.sysml.runtime.util:org.apache.sysml.utils:org.apache.sysml.yarn</excludePackageNames>\n<additionalparam>${javadoc.opts}</additionalparam>\n</configuration>\n<descriptors>\n<descriptor>src/assembly/lite.xml</descriptor>\n</descriptors>\n+ <archive>\n+ <manifest>\n+ <mainClass>org.apache.sysml.api.DMLScript</mainClass>\n+ </manifest>\n+ <manifestEntries>\n+ <Build-Time>${maven.build.timestamp}</Build-Time>\n+ <Group-Id>${project.groupId}</Group-Id>\n+ <Artifact-Id>${project.artifactId}-lite</Artifact-Id>\n+ <Version>${project.version}</Version>\n+ </manifestEntries>\n+ </archive>\n</configuration>\n</execution>\n</executions>\n</profile>\n<profile>\n- <!-- Profile to create standalone jar.\n- Execute with `mvn clean package -P standalone-jar` -->\n+ <!-- Profile to create standalone jar. Execute with `mvn clean package\n+ -P standalone-jar` -->\n<id>standalone-jar</id>\n<build>\n<plugins>\n" }, { "change_type": "MODIFY", "old_path": "src/assembly/lite.xml", "new_path": "src/assembly/lite.xml", "diff": "<dependencySets>\n<dependencySet>\n<includes>\n- <include>*:antlr4-runtime</include>\n+ <include>*:antlr4</include>\n</includes>\n<unpackOptions>\n<includes>\n</includes>\n<unpackOptions>\n<includes>\n+ <include>org/apache/commons/cli/OptionValidator.class</include>\n+ <include>org/apache/commons/cli/Util.class</include>\n<include>org/apache/commons/cli/AlreadySelectedException.class</include>\n<include>org/apache/commons/cli/CommandLine.class</include>\n<include>org/apache/commons/cli/CommandLineParser.class</include>\n<unpackOptions>\n<includes>\n<include>META-INF/services/org.apache.hadoop.fs.FileSystem</include>\n+ <include>common-version-info.properties</include>\n<include>org/apache/hadoop/log/metrics/EventCounter.class</include>\n+ <include>org/apache/hadoop/security/Groups$CachedGroups.class</include>\n+ <include>org/apache/hadoop/security/token/Token.class</include>\n+ <include>org/apache/hadoop/security/token/TokenIdentifier.class</include>\n+ <include>org/apache/hadoop/util/GenericOptionsParser.class</include>\n<include>org/apache/hadoop/util/ShutdownHookManager$2.class</include>\n<include>org/apache/hadoop/HadoopIllegalArgumentException.class</include>\n<include>org/apache/hadoop/conf/Configurable.class</include>\n<unpack>true</unpack>\n</dependencySet>\n+ <dependencySet>\n+ <includes>\n+ <include>*:hadoop-mapreduce-client-jobclient</include>\n+ </includes>\n+ <unpackOptions>\n+ <includes>\n+ <include>org/apache/hadoop/mapred/YarnClientProtocolProvider.class</include>\n+ </includes>\n+ </unpackOptions>\n+ <scope>compile</scope>\n+ <unpack>true</unpack>\n+ </dependencySet>\n+\n<dependencySet>\n<includes>\n<include>*:jcl-over-slf4j</include>\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/utils/lite/BuildLite.java", "new_path": "src/main/java/org/apache/sysml/utils/lite/BuildLite.java", "diff": "@@ -92,6 +92,14 @@ public class BuildLite {\nadditionalResources.add(\"org/apache/hadoop/util/ShutdownHookManager$2.class\");\nadditionalResources.add(\"org/apache/hadoop/log/metrics/EventCounter.class\");\n+ additionalResources.add(\"org/apache/hadoop/util/GenericOptionsParser.class\");\n+ additionalResources.add(\"org/apache/hadoop/security/token/Token.class\");\n+ additionalResources.add(\"org/apache/hadoop/security/token/TokenIdentifier.class\");\n+ additionalResources.add(\"org/apache/hadoop/security/Groups$CachedGroups.class\");\n+ additionalResources.add(\"org/apache/commons/cli/OptionValidator.class\");\n+ additionalResources.add(\"org/apache/commons/cli/Util.class\");\n+ additionalResources.add(\"common-version-info.properties\");\n+\n}\n/**\n@@ -104,7 +112,17 @@ public class BuildLite {\nhadoopCommonResources.add(\"META-INF/services/org.apache.hadoop.fs.FileSystem\");\nhadoopCommonResources.add(\"org/apache/hadoop/util/ShutdownHookManager$2.class\");\nhadoopCommonResources.add(\"org/apache/hadoop/log/metrics/EventCounter.class\");\n+ hadoopCommonResources.add(\"org/apache/hadoop/util/GenericOptionsParser.class\");\n+ hadoopCommonResources.add(\"org/apache/hadoop/security/token/Token.class\");\n+ hadoopCommonResources.add(\"org/apache/hadoop/security/token/TokenIdentifier.class\");\n+ hadoopCommonResources.add(\"org/apache/hadoop/security/Groups$CachedGroups.class\");\n+ hadoopCommonResources.add(\"common-version-info.properties\");\nadditionalJarToFileMappingsForDependencySets.put(\"hadoop-common\", hadoopCommonResources);\n+\n+ SortedSet<String> commonsCliResources = new TreeSet<String>();\n+ commonsCliResources.add(\"org/apache/commons/cli/OptionValidator.class\");\n+ commonsCliResources.add(\"org/apache/commons/cli/Util.class\");\n+ additionalJarToFileMappingsForDependencySets.put(\"commons-cli\", commonsCliResources);\n}\n/**\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/utils/lite/BuildLiteExecution.java", "new_path": "src/main/java/org/apache/sysml/utils/lite/BuildLiteExecution.java", "diff": "@@ -46,7 +46,8 @@ import org.apache.sysml.runtime.util.DataConverter;\n*/\npublic class BuildLiteExecution {\n- protected static Logger log = Logger.getLogger(BuildLiteExecution.class);\n+ private static Logger log = Logger.getLogger(BuildLiteExecution.class);\n+ private static final String ROOT = \"functions/jmlc/temp/\";\npublic static void main(String[] args) throws Exception {\n@@ -94,11 +95,11 @@ public class BuildLiteExecution {\n\"write(predicted_y, \\\"./tmp\\\", format=\\\"text\\\");\\n\";\n/* @formatter:on */\n- File file = new File(\"temp/scoring-example.dml\");\n+ File file = new File(ROOT+\"scoring-example.dml\");\nFileUtils.writeStringToFile(file, scriptString);\nConnection conn = getConfiguredConnection();\n- String dml = conn.readScript(\"temp/scoring-example.dml\");\n+ String dml = conn.readScript(ROOT+\"scoring-example.dml\");\nPreparedScript script = conn.prepareScript(dml, new String[] { \"W\", \"X\" }, new String[] { \"predicted_y\" },\nfalse);\n@@ -158,17 +159,17 @@ public class BuildLiteExecution {\npublic static void jmlcWriteMatrix() throws Exception {\nConnection conn = getConfiguredConnection();\nPreparedScript script = conn.prepareScript(\n- \"x=matrix('1 2 3 4',rows=2,cols=2);write(x,'temp/x.csv',format='csv');\", new String[] {},\n+ \"x=matrix('1 2 3 4',rows=2,cols=2);write(x,'\"+ROOT+\"x.csv',format='csv');\", new String[] {},\nnew String[] {}, false);\nscript.executeScript();\n/* @formatter:off */\nString scriptString =\n\"m = matrix('1 2 3 0 0 0 7 8 9 0 0 0', rows=4, cols=3)\\n\" +\n- \"write(m, 'temp/m.txt', format='text')\\n\" +\n- \"write(m, 'temp/m.mm', format='mm')\\n\" +\n- \"write(m, 'temp/m.csv', format='csv')\\n\" +\n- \"write(m, 'temp/m.binary', format='binary')\\n\";\n+ \"write(m, '\"+ROOT+\"m.txt', format='text')\\n\" +\n+ \"write(m, '\"+ROOT+\"m.mm', format='mm')\\n\" +\n+ \"write(m, '\"+ROOT+\"m.csv', format='csv')\\n\" +\n+ \"write(m, '\"+ROOT+\"m.binary', format='binary')\\n\";\n/* @formatter:on */\nscript = conn.prepareScript(scriptString, new String[] {}, new String[] {}, false);\n@@ -179,12 +180,12 @@ public class BuildLiteExecution {\npublic static void jmlcReadMatrix() throws Exception {\nConnection conn = getConfiguredConnection();\n- PreparedScript script = conn.prepareScript(\"x=read('temp/x.csv',format='csv');y=x*2;print(toString(y));\",\n+ PreparedScript script = conn.prepareScript(\"x=read('\"+ROOT+\"x.csv',format='csv');y=x*2;print(toString(y));\",\nnew String[] {}, new String[] {}, false);\nscript.executeScript();\n/* @formatter:off */\n- String scriptString = \"m = read('temp/m.csv',format='csv')\\n\" +\n+ String scriptString = \"m = read('\"+ROOT+\"m.csv',format='csv')\\n\" +\n\"print(toString(m))\\n\" +\n\"print('min:' + min(m))\\n\" +\n\"print('max:' + max(m))\\n\" +\n@@ -201,9 +202,9 @@ public class BuildLiteExecution {\n// note: the following can be set to work using the following setting\n// in the Connection class: cconf.set(ConfigType.IGNORE_READ_WRITE_METADATA, false);\n- // \"m2=read('temp/m.txt', format='text')\\n\" +\n- // \"m3=read('temp/m.mm', format='mm')\\n\" +\n- // \"m4=read('temp/m.binary', format='binary')\\n\" +\n+ // \"m2=read('\"+ROOT+\"m.txt', format='text')\\n\" +\n+ // \"m3=read('\"+ROOT+\"m.mm', format='mm')\\n\" +\n+ // \"m4=read('\"+ROOT+\"m.binary', format='binary')\\n\" +\n// \"print('m2:'+toString(m2))\\n\" +\n// \"print('m3:'+toString(m3))\\n\" +\n// \"print('m4:'+toString(m4))\\n\";\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/jmlc/BuildLiteJarTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/jmlc/BuildLiteJarTest.java", "diff": "@@ -46,12 +46,8 @@ public class BuildLiteJarTest extends AutomatedTestBase\n}\n@Test\n- public void testJMLCWriteMatrix() throws Exception {\n+ public void testJMLCWriteReadMatrix() throws Exception {\nBuildLiteExecution.jmlcWriteMatrix();\n- }\n-\n- @Test\n- public void testJMLCReadMatrix() throws Exception {\nBuildLiteExecution.jmlcReadMatrix();\n}\n" }, { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/jmlc/MulticlassSVMScoreTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/jmlc/MulticlassSVMScoreTest.java", "diff": "@@ -181,12 +181,8 @@ public class MulticlassSVMScoreTest extends AutomatedTestBase\nprivate ArrayList<double[][]> generateInputs( int num, int rows, int cols, double sparsity )\n{\nArrayList<double[][]> ret = new ArrayList<double[][]>();\n-\nfor( int i=0; i<num; i++ )\n- {\n- double[][] X = getRandomMatrix(rows, cols, -1, 1, sparsity, System.nanoTime());\n- ret.add(X);\n- }\n+ ret.add(getRandomMatrix(rows, cols, -1, 1, sparsity, i));\nreturn ret;\n}\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixReorg.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixReorg.java", "diff": "@@ -445,7 +445,11 @@ public class LibMatrixReorg\n//check for same dimensions\nif( rlen==rows && clen == cols ) {\n- out.copy(in); //incl dims, nnz\n+ //copy incl dims, nnz\n+ if( SHALLOW_COPY_REORG )\n+ out.copyShallow(in);\n+ else\n+ out.copy(in);\nreturn out;\n}\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java", "new_path": "src/main/java/org/apache/sysml/runtime/matrix/data/MatrixBlock.java", "diff": "@@ -761,17 +761,15 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab\n}\nelse //SPARSE <- DENSE\n{\n- for( int i=0; i<that.rlen; i++ )\n- {\n- int aix = rowoffset+i;\n- for( int j=0, bix=i*that.clen; j<that.clen; j++ )\n- {\n- double val = that.denseBlock[bix+j];\n- if( val != 0 ) {\n- //create sparserow only if required\n+ double[] b = that.denseBlock;\n+ final int bm = that.rlen;\n+ final int bn = that.clen;\n+ for( int i=0, aix=rowoffset, bix=0; i<bm; i++, aix++, bix+=bn )\n+ for( int j=0; j<bn; j++ ) {\n+ final double bval = b[bix+j];\n+ if( bval != 0 ) {\nsparseBlock.allocate(aix, estimatedNNzsPerRow, clen);\n- sparseBlock.append(aix, coloffset+j, val);\n- }\n+ sparseBlock.append(aix, coloffset+j, bval);\n}\n}\n}\n@@ -1351,6 +1349,17 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab\ncopyDenseToDense(that);\n}\n+ public void copyShallow(MatrixBlock that) {\n+ rlen = that.rlen;\n+ clen = that.clen;\n+ nonZeros = that.nonZeros;\n+ sparse = that.sparse;\n+ if( !sparse )\n+ denseBlock = that.denseBlock;\n+ else\n+ sparseBlock = that.sparseBlock;\n+ }\n+\nprivate void copySparseToSparse(MatrixBlock that)\n{\nthis.nonZeros=that.nonZeros;\n@@ -3546,7 +3555,7 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab\n//core append operation\n//copy left and right input into output\n- if( !result.sparse ) //DENSE\n+ if( !result.sparse && nnz!=0 ) //DENSE\n{\nif( cbind ) {\nresult.copy(0, m-1, 0, clen-1, this, false);\n@@ -3563,13 +3572,12 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab\n}\n}\n}\n- else //SPARSE\n+ else if(nnz != 0) //SPARSE\n{\n//adjust sparse rows if required\n- if( !this.isEmptyBlock(false) || !Arrays.stream(that).allMatch(mb -> mb.isEmptyBlock(false)) ) {\nresult.allocateSparseRowsBlock();\n//allocate sparse rows once for cbind\n- if( cbind && result.getSparseBlock() instanceof SparseBlockMCSR ) {\n+ if( cbind && nnz > rlen && result.getSparseBlock() instanceof SparseBlockMCSR ) {\nSparseBlock sblock = result.getSparseBlock();\nfor( int i=0; i<result.rlen; i++ ) {\nfinal int row = i; //workaround for lambda compile issue\n@@ -3578,7 +3586,6 @@ public class MatrixBlock extends MatrixValue implements CacheBlock, Externalizab\nsblock.allocate(i, lnnz);\n}\n}\n- }\n//core append operation\nresult.appendToSparse(this, 0, 0);\n" } ]

[ { "change_type": "MODIFY", "old_path": "docs/release-process.md", "new_path": "docs/release-process.md", "diff": "@@ -139,9 +139,7 @@ Verify that the snapshot is now available at\n# Release Candidate Build and Deployment\n-To be written. (Describe how the release candidate is built, including checksums. Describe how\n-the release candidate is deployed to servers for review.)\n-\n+For detailed information, please see [SystemML Release Creation Process](release-creation-process.html).\n# Release Candidate Checklist\n@@ -488,4 +486,3 @@ Commit the update to `documentation.html` to publish the website update.\nThe versioned project documentation is now deployed to the main website, and the\n[Documentation Page](http://systemml.apache.org/documentation) contains a link to the versioned documentation.\n-\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateCell.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateCell.java", "diff": "@@ -75,8 +75,8 @@ public class TemplateCell extends TemplateBase\npublic boolean open(Hop hop) {\nreturn hop.dimsKnown() && isValidOperation(hop)\n&& !(hop.getDim1()==1 && hop.getDim2()==1)\n- || (hop instanceof IndexingOp && (((IndexingOp)hop)\n- .isColLowerEqualsUpper() || hop.getDim2()==1));\n+ || (hop instanceof IndexingOp && hop.getInput().get(0).getDim2() > 0\n+ && (((IndexingOp)hop).isColLowerEqualsUpper() || hop.getDim2()==1));\n}\n@Override\n" }, { "change_type": "MODIFY", "old_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateRow.java", "new_path": "src/main/java/org/apache/sysml/hops/codegen/template/TemplateRow.java", "diff": "@@ -95,7 +95,8 @@ public class TemplateRow extends TemplateBase\n|| (hop instanceof AggUnaryOp && ((AggUnaryOp)hop).getDirection()!=Direction.RowCol\n&& hop.getInput().get(0).getDim1()>1 && hop.getInput().get(0).getDim2()>1\n&& HopRewriteUtils.isAggUnaryOp(hop, SUPPORTED_ROW_AGG))\n- || (hop instanceof IndexingOp && HopRewriteUtils.isColumnRangeIndexing((IndexingOp)hop));\n+ || (hop instanceof IndexingOp && hop.getInput().get(0).getDim2() > 0\n+ && HopRewriteUtils.isColumnRangeIndexing((IndexingOp)hop));\n}\n@Override\n@@ -449,7 +450,7 @@ public class TemplateRow extends TemplateBase\nout = new CNodeTernary(cdata1,\nTemplateUtils.createCNodeData(new LiteralOp(hop.getInput().get(0).getDim2()), true),\nTemplateUtils.createCNodeData(hop.getInput().get(4), true),\n- (!hop.dimsKnown()||hop.getDim2()>1) ? TernaryType.LOOKUP_RVECT1 : TernaryType.LOOKUP_RC1);\n+ (hop.getDim2() != 1) ? TernaryType.LOOKUP_RVECT1 : TernaryType.LOOKUP_RC1);\n}\nif( out == null ) {\n" } ]

[ { "change_type": "MODIFY", "old_path": "src/test/java/org/apache/sysml/test/integration/functions/jmlc/MulticlassSVMScoreTest.java", "new_path": "src/test/java/org/apache/sysml/test/integration/functions/jmlc/MulticlassSVMScoreTest.java", "diff": "@@ -39,7 +39,6 @@ import org.junit.Test;\npublic class MulticlassSVMScoreTest extends AutomatedTestBase\n{\n-\nprivate final static String TEST_NAME = \"m-svm-score\";\nprivate final static String TEST_DIR = \"functions/jmlc/\";\nprivate final static String MODEL_FILE = \"sentiment_model.mtx\";\n@@ -49,15 +48,17 @@ public class MulticlassSVMScoreTest extends AutomatedTestBase\nprivate final static int rows = 107;\nprivate final static int cols = 46; //fixed\n- private final static int nRuns = 5;\n+ private final static int nRuns = 3;\nprivate final static double sparsity1 = 0.7;\nprivate final static double sparsity2 = 0.1;\n+ //This testcase recently caused intermittent test failures on jenkins that are not\n+ //reproducible in local environments; hence we perform additional sanity checks here.\n+ private final static boolean CHECK_IN_OUT = true;\n@Override\n- public void setUp()\n- {\n+ public void setUp() {\naddTestConfiguration(TEST_NAME, new TestConfiguration(TEST_CLASS_DIR, TEST_NAME, new String[] { \"predicted_y\" }) );\n}\n@@ -89,9 +90,13 @@ public class MulticlassSVMScoreTest extends AutomatedTestBase\n//generate inputs\nArrayList<double[][]> Xset = generateInputs(nRuns, rows, cols, sparse?sparsity2:sparsity1);\n+ if( CHECK_IN_OUT )\n+ checkSelfEquivalence(Xset, rows, cols);\n//run DML via JMLC\nArrayList<double[][]> Yset = execDMLScriptviaJMLC( Xset, flags );\n+ if( CHECK_IN_OUT )\n+ checkSelfEquivalence(Yset, rows, 1);\n//run R and compare results to DML result\nString HOME = SCRIPT_DIR + TEST_DIR;\n@@ -162,13 +167,11 @@ public class MulticlassSVMScoreTest extends AutomatedTestBase\nret.add(Y); //keep result for comparison\n}\n}\n- catch(Exception ex)\n- {\n+ catch(Exception ex) {\nex.printStackTrace();\nthrow new IOException(ex);\n}\n- finally\n- {\n+ finally {\nif( conn != null )\nconn.close();\n}\n@@ -178,13 +181,18 @@ public class MulticlassSVMScoreTest extends AutomatedTestBase\nreturn ret;\n}\n- private ArrayList<double[][]> generateInputs( int num, int rows, int cols, double sparsity )\n- {\n+ private ArrayList<double[][]> generateInputs( int num, int rows, int cols, double sparsity ) {\nArrayList<double[][]> ret = new ArrayList<double[][]>();\nfor( int i=0; i<num; i++ )\n- ret.add(getRandomMatrix(rows, cols, -1, 1, sparsity, i));\n-\n+ ret.add(getRandomMatrix(rows, cols, -1, 1, sparsity, 7));\nreturn ret;\n}\n+ private void checkSelfEquivalence(ArrayList<double[][]> data, int rows, int cols) {\n+ if( data == null || data.size() < 2 )\n+ return;\n+ double[][] data0 = data.get(0);\n+ for(int i=1; i<data.size(); i++)\n+ TestUtils.compareMatrices(data0, data.get(i), rows, cols, eps);\n+ }\n}\n\\ No newline at end of file\n" } ]