ku-te.py

from typing import Dict

import datasets as ds
import pandas as pd

_CITATION = """\
@inproceedings{,
    author = "小谷通隆 and 柴田知秀 and 中田貴之 and 黒橋禎夫",
    title = "日本語Textual Entailmentのデータ構築と自動獲得した類義表現に基づく推論関係の認識",
    booktitle = "言語処理学会第14回年次大会",
    year = "2008",
    url = "https://nlp.ist.i.kyoto-u.ac.jp/?Textual+Entailment+%E8%A9%95%E4%BE%A1%E3%83%87%E3%83%BC%E3%82%BF",
    pages = "1140-1143"
}
"""

_DESCRIPTION = """\

"""

_HOMEPAGE = "https://nlp.ist.i.kyoto-u.ac.jp/?Textual+Entailment+%E8%A9%95%E4%BE%A1%E3%83%87%E3%83%BC%E3%82%BF"

_LICENSE = "unknown"

_URL = "http://nlp.ist.i.kyoto-u.ac.jp/DLcounter/lime.cgi?down=http://nlp.ist.i.kyoto-u.ac.jp/nl-resource/rte/entail_evaluation_set.txt&name=entail_evaluation_set.txt"


class KUTEDataset(ds.GeneratorBasedBuilder):
    VERSION = ds.Version("1.0.0")
    DEFAULT_CONFIG_NAME = "3way"

    BUILDER_CONFIGS = [
        ds.BuilderConfig(
            name="original",
            version=VERSION,
            description="hoge",
        ),
        ds.BuilderConfig(
            name="3way",
            version=VERSION,
            description="fuga",
        ),
        ds.BuilderConfig(
            name="3way-strict",
            version=VERSION,
            description="fuga",
        ),
        ds.BuilderConfig(
            name="2way",
            version=VERSION,
            description="fuga",
        ),
    ]

    def _info(self) -> ds.DatasetInfo:
        if self.config.name == "original":
            labels = ds.ClassLabel(names=["◎", "○", "△", "×"])
        elif self.config.name in ["3way", "3way-strict"]:
            labels = ds.ClassLabel(names=["entailment", "neutral", "contradiction"])
        elif self.config.name == "2way":
            labels = ds.ClassLabel(names=["entailment", "non-entailment"])

        features = ds.Features(
            {
                "id": ds.Value("int32"),
                "premise": ds.Value("string"),
                "hypothesis": ds.Value("string"),
                "label": labels,
                "category_subcategory": ds.Value("string"),
            }
        )
        return ds.DatasetInfo(
            description=_DESCRIPTION,
            citation=_CITATION,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            features=features,
        )

    def _split_generators(self, dl_manager: ds.DownloadManager):
        data_path = dl_manager.download_and_extract(_URL)
        df: pd.DataFrame = pd.read_table(data_path, sep=" ", header=None)
        df.columns = ["id", "category_subcategory", "label", "premise", "hypothesis"]

        if self.config.name == "original":
            mapping = None
        elif self.config.name == "3way":
            mapping = {
                "◎": "entailment",
                "○": "entailment",
                "△": "neutral",
                "×": "contradiction",
            }
        elif self.config.name == "3way-strict":
            mapping = {
                "◎": "entailment",
                "○": None,
                "△": "neutral",
                "×": "contradiction",
            }
        elif self.config.name == "2way":
            mapping = {
                "◎": "entailment",
                "○": "entailment",
                "△": "non-entailment",
                "×": "non-entailment",
            }

        return [
            ds.SplitGenerator(
                name=ds.Split.TEST,
                gen_kwargs={"df": df, "mapping": mapping},
            ),
        ]

    def _generate_examples(self, df: pd.DataFrame, mapping: Dict[str, str] = None):
        data = []
        for row in df.to_dict("records"):
            if mapping is not None:
                row["label"] = mapping[row["label"]]
            if row["label"] is None:
                continue
            data.append(row)

        for i, row in enumerate(data):
            yield i, row