PaddlePaddle/dureader_robust

1# coding=utf-82# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.3# Copyright 2020 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors.4#5# Licensed under the Apache License, Version 2.0 (the "License");6# you may not use this file except in compliance with the License.7# You may obtain a copy of the License at8#9#     http://www.apache.org/licenses/LICENSE-2.010#11# Unless required by applicable law or agreed to in writing, software12# distributed under the License is distributed on an "AS IS" BASIS,13# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.14# See the License for the specific language governing permissions and15# limitations under the License.16 17# Lint as: python318 19 20import json21import os22 23import datasets24 25 26logger = datasets.logging.get_logger(__name__)27 28_DESCRIPTION = """\29DureaderRobust is a chinese reading comprehension \30dataset, designed to evaluate the MRC models from \31three aspects: over-sensitivity, over-stability \32and generalization.33"""34 35_URL = "https://bj.bcebos.com/paddlenlp/datasets/dureader_robust-data.tar.gz"36 37 38class DureaderRobustConfig(datasets.BuilderConfig):39    """BuilderConfig for DureaderRobust."""40 41    def __init__(self, **kwargs):42        """BuilderConfig for DureaderRobust.43 44        Args:45          **kwargs: keyword arguments forwarded to super.46        """47        super(DureaderRobustConfig, self).__init__(**kwargs)48 49 50class DureaderRobust(datasets.GeneratorBasedBuilder):51    BUILDER_CONFIGS = [52        DureaderRobustConfig(53            name="plain_text",54            version=datasets.Version("1.0.0", ""),55            description="Plain text",56        ),57    ]58 59    def _info(self):60        return datasets.DatasetInfo(61            description=_DESCRIPTION,62            features=datasets.Features(63                {64                    "id": datasets.Value("string"),65                    "title": datasets.Value("string"),66                    "context": datasets.Value("string"),67                    "question": datasets.Value("string"),68                    "answers": datasets.features.Sequence(69                        {70                            "text": datasets.Value("string"),71                            "answer_start": datasets.Value("int32"),72                        }73                    ),74                }75            ),76            # No default supervised_keys (as we have to pass both question77            # and context as input).78            supervised_keys=None,79            homepage="https://arxiv.org/abs/2004.11142",80        )81 82    def _split_generators(self, dl_manager):83        dl_dir = dl_manager.download_and_extract(_URL)84 85        return [86            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": os.path.join(dl_dir,'dureader_robust-data', 'train.json')}),87            datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": os.path.join(dl_dir,'dureader_robust-data', 'dev.json')}),88            datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": os.path.join(dl_dir,'dureader_robust-data', 'test.json')}),89        ]90 91    def _generate_examples(self, filepath):92        """This function returns the examples in the raw (text) form."""93        logger.info("generating examples from = %s", filepath)94        key = 095        with open(filepath, encoding="utf-8") as f:96            durobust = json.load(f)97            for article in durobust["data"]:98                title = article.get("title", "")99                for paragraph in article["paragraphs"]:100                    context = paragraph["context"]  # do not strip leading blank spaces GH-2585101                    for qa in paragraph["qas"]:102                        answer_starts = [answer["answer_start"] for answer in qa.get("answers",'')]103                        answers = [answer["text"] for answer in qa.get("answers",'')]104                        # Features currently used are "context", "question", and "answers".105                        # Others are extracted here for the ease of future expansions.106                        yield key, {107                            "title": title,108                            "context": context,109                            "question": qa["question"],110                            "id": qa["id"],111                            "answers": {112                                "answer_start": answer_starts,113                                "text": answers,114                            },115                        }116                        key += 1117