Create simclue.py

simclue.py

@@ -0,0 +1,124 @@
+# -*- coding: utf-8 -*-
+"""
+@author:XuMing([email protected])
+@description:
+"""
+
+"""Natural Language Inference (NLI) Chinese Corpus.(nli_zh)"""
+
+import os
+import json
+import datasets
+
+_DESCRIPTION = """SimCLUE：3000000+中文语义理解与匹配数据集"""
+
+GITHUB_HOME = "https://github.com/CLUEbenchmark/SimCLUE"
+
+_CITATION = "https://github.com/CLUEbenchmark/SimCLUE"
+
+_DATA_URL = "https://storage.googleapis.com/cluebenchmark/tasks/simclue_public.zip"
+
+
+class SimCLUEConfig(datasets.BuilderConfig):
+
+    def __init__(self, features, data_url, citation, url, label_classes=(0, 1), **kwargs):
+        super().__init__(version=datasets.Version("1.0.0"), **kwargs)
+        self.features = features
+        self.label_classes = label_classes
+        self.data_url = data_url
+        self.citation = citation
+        self.url = url
+
+
+class SimCLUE(datasets.GeneratorBasedBuilder):
+    """The Natural Language Inference Chinese(NLI_zh) Corpus."""
+
+    part_file = {'train_rank': 'train_rank.json',
+                 'train_pair': 'train_pair.json',
+                 'corpus': 'corpus.txt',
+                 'train_pair_postive': 'train_pair_postive.json',
+                 'dev': 'dev.json',
+                 'test_public': 'test_public.json'}
+
+    BUILDER_CONFIGS = [
+        SimCLUEConfig(
+            name="train_rank",
+            description=_DESCRIPTION,
+            features=datasets.Features({"query": datasets.Value("string"),
+                                          "title": datasets.Value("string"),
+                                          "neg_title": datasets.Value("string")}),
+            data_url=_DATA_URL,
+            citation=_CITATION,
+            url=GITHUB_HOME,
+        ),
+        SimCLUEConfig(
+            name="train_pair",
+            description=_DESCRIPTION,
+            features=datasets.Features({"sentence1": datasets.Value("string"),
+                                       "sentence2": datasets.Value("string"),
+                                       "label": datasets.Value("int32")}),
+            data_url=_DATA_URL,
+            citation=_CITATION,
+            url=GITHUB_HOME,
+        ),
+        SimCLUEConfig(
+            name="corpus",
+            description=_DESCRIPTION,
+            features=datasets.Features({"sentence1": datasets.Value("string")}),
+            data_url=_DATA_URL,
+            citation=_CITATION,
+            url=GITHUB_HOME,
+        ),
+        SimCLUEConfig(
+            name="train_pair_postive",
+            description=_DESCRIPTION,
+            features=datasets.Features({"sentence1": datasets.Value("string"),
+                                          "sentence2": datasets.Value("string"),
+                                          "label": datasets.Value("int32")}),
+            data_url=_DATA_URL,
+            citation=_CITATION,
+            url=GITHUB_HOME,
+        ),
+        SimCLUEConfig(
+            name="dev",
+            description=_DESCRIPTION,
+            features=datasets.Features({"sentence1": datasets.Value("string"),
+                                          "sentence2": datasets.Value("string"),
+                                          "label": datasets.Value("int32")}),
+            data_url=_DATA_URL,
+            citation=_CITATION,
+            url=GITHUB_HOME,
+        ),
+        SimCLUEConfig(
+            name="test_public",
+            description=_DESCRIPTION,
+            features=datasets.Features({"sentence1": datasets.Value("string"),
+                                          "sentence2": datasets.Value("string"),
+                                          "label": datasets.Value("int32")}),
+            data_url=_DATA_URL,
+            citation=_CITATION,
+            url=GITHUB_HOME,
+        ),
+    ]
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=self.config.description,
+            features=self.config.features,
+            homepage=self.config.url,
+            citation=self.config.citation,
+        )
+
+    def _split_generators(self, dl_manager):
+        dl_dir = dl_manager.download_and_extract(self.config.data_url)
+        return [datasets.SplitGenerator(
+                        name=self.config.name,
+                        gen_kwargs={
+                            "filepath": os.path.join(dl_dir, self.part_file[self.config.name]),
+                        })]
+
+    def _generate_examples(self, filepath):
+        """This function returns the examples in the raw (text) form."""
+        with open(filepath, 'r', encoding="utf-8") as f:
+            for idx, row in enumerate(f):
+                yield idx, json.loads(row)