qg_squadshifts.py

""" python -c "from datasets import load_dataset;load_dataset('.')" """
import json
from itertools import chain
import datasets

logger = datasets.logging.get_logger(__name__)
_DESCRIPTION = """[SQuAD Shifts](https://modestyachts.github.io/squadshifts-website/index.html) dataset for question generation (QG) task."""
_URL = 'https://huggingface.co/datasets/asahi417/qg_squadshift/raw/main/data/processed'
_FILES = {
    str(datasets.Split.TEST): {
            'new_wiki': [f'{_URL}/new_wiki.test{i:02d}.jsonl' for i in range(3)],
            'nyt': [f'{_URL}/nyt.test{i:02d}.jsonl' for i in range(4)],
            'reddit': [f'{_URL}/reddit.test{i:02d}.jsonl' for i in range(4)],
            'amazon': [f'{_URL}/amazon.test{i:02d}.jsonl' for i in range(4)]
        },
    str(datasets.Split.TRAIN): {
            'new_wiki': [f'{_URL}/new_wiki.train{i:02d}.jsonl' for i in range(2)],
            'nyt': [f'{_URL}/nyt.train{i:02d}.jsonl' for i in range(3)],
            'reddit': [f'{_URL}/reddit.train{i:02d}.jsonl' for i in range(3)],
            'amazon': [f'{_URL}/amazon.train{i:02d}.jsonl' for i in range(3)]
        },
    str(datasets.Split.VALIDATION): {
            'new_wiki': [f'{_URL}/new_wiki.validation{i:02d}.jsonl' for i in range(2)],
            'nyt': [f'{_URL}/nyt.validation{i:02d}.jsonl' for i in range(3)],
            'reddit': [f'{_URL}/reddit.validation{i:02d}.jsonl' for i in range(3)],
            'amazon': [f'{_URL}/amazon.validation{i:02d}.jsonl' for i in range(3)]
        },
}
# _FILES = {
#     str(datasets.Split.TEST): {
#             'new_wiki': [f'{_URL}/new_wiki.test.jsonl'],
#             'nyt': [f'{_URL}/nyt.test.jsonl'],
#             'reddit': [f'{_URL}/reddit.test.jsonl'],
#             'amazon': [f'{_URL}/amazon.test.jsonl']
#         },
#     str(datasets.Split.TRAIN): {
#             'new_wiki': [f'{_URL}/new_wiki.train.jsonl'],
#             'nyt': [f'{_URL}/nyt.train.jsonl'],
#             'reddit': [f'{_URL}/reddit.train.jsonl'],
#             'amazon': [f'{_URL}/amazon.train.jsonl']
#         },
#     str(datasets.Split.VALIDATION): {
#             'new_wiki': [f'{_URL}/new_wiki.validation.jsonl'],
#             'nyt': [f'{_URL}/nyt.validation.jsonl'],
#             'reddit': [f'{_URL}/reddit.validation.jsonl'],
#             'amazon': [f'{_URL}/amazon.validation.jsonl']
#         },
# }
_DOMAIN = list(_FILES[list(_FILES.keys())[0]].keys())


class QGSQuADShiftsConfig(datasets.BuilderConfig):
    """BuilderConfig for SquadQG"""

    def __init__(self, **kwargs):
        """BuilderConfig for SquadQG.
        Args:
          **kwargs: keyword arguments forwarded to super.
        """
        super(QGSQuADShiftsConfig, self).__init__(**kwargs)


class QGSQuADShifts(datasets.GeneratorBasedBuilder):

    BUILDER_CONFIGS = [QGSQuADShiftsConfig(name="default", description="All domain.")]
    BUILDER_CONFIGS += [QGSQuADShiftsConfig(name=i, description=f"Domain {i}") for i in sorted(_DOMAIN)]

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "answer": datasets.Value("string"),
                    "question": datasets.Value("string"),
                    "sentence": datasets.Value("string"),
                    "paragraph": datasets.Value("string"),
                    "sentence_answer": datasets.Value("string"),
                    "paragraph_answer": datasets.Value("string"),
                    "paragraph_sentence": datasets.Value("string"),
                    "paragraph_id": datasets.Value("string")
                }
            ),
            supervised_keys=None,
            homepage="https://github.com/asahi417/lm-question-generation"
        )

    def _split_generators(self, dl_manager):
        if self.config.name == 'default':
            downloaded_file = dl_manager.download_and_extract({k: list(chain(*list(v.values()))) for k, v in _FILES.items()})
        else:
            downloaded_file = dl_manager.download_and_extract({k: v[self.config.name] for k, v in _FILES.items()})
        return [datasets.SplitGenerator(name=k, gen_kwargs={"filepaths": downloaded_file[k]}) for k in _FILES.keys()]

    def _generate_examples(self, filepaths):
        _key = 0
        for filepath in filepaths:
            logger.info("generating examples from = %s", filepath)
            with open(filepath, encoding="utf-8") as f:
                _list = f.read().split('\n')
                if _list[-1] == '':
                    _list = _list[:-1]
                for i in _list:
                    data = json.loads(i)
                    yield _key, data
                    _key += 1