nlprepl.py

"""NKJP1M: The manually annotated subcorpus of the National Corpus of Polish"""
import conllu
import datasets


logger = datasets.logging.get_logger(__name__)


_CITATION = ""

BY_NAME = "by_name"
BY_TYPE = "by_type"

TAGSET_NKJP = "nkjp"
TAGSET_UD = "ud"

_DESCRIPTION = {
    BY_NAME: {
        TAGSET_NKJP: "NLPrePL divided by document name for NKJP tagset",
        TAGSET_UD: "TNLPrePL divided by document name for UD tagset"
    },
    BY_TYPE: {
        TAGSET_NKJP: "NLPrePL divided by document type for NKJP tagset",
        TAGSET_UD: "NLPrePL divided by document type for UD tagset"
    }
}

_TYPES = [BY_NAME, BY_TYPE]
_TAGSETS = [TAGSET_NKJP, TAGSET_UD]

_URLS = {
    BY_NAME: {
        TAGSET_NKJP: {
            'train': "nkjp_tagset/fair_by_document_name/train_nlprepl-nkjp.conllu.gz",
            'dev': "nkjp_tagset/fair_by_document_name/dev_nlprepl-nkjp.conllu.gz",
            'test': "nkjp_tagset/fair_by_document_name/test_nlprepl-nkjp.conllu.gz"
        },
        TAGSET_UD: {
            'train': "ud_tagset/fair_by_document_name/train_nlprepl-ud.conllu.gz",
            'dev': "ud_tagset/fair_by_document_name/dev_nlprepl-ud.conllu.gz",
            'test': "ud_tagset/fair_by_document_name/test_nlprepl-ud.conllu.gz"
        }
    },
    BY_TYPE: {
        TAGSET_NKJP: {
            'train': "nkjp_tagset/fair_by_document_type/train_nlprepl-nkjp.conllu.gz",
            'dev': "nkjp_tagset/fair_by_document_type/dev_nlprepl-nkjp.conllu.gz",
            'test': "nkjp_tagset/fair_by_document_type/test_nlprepl-nkjp.conllu.gz"
        },
        TAGSET_UD: {
            'train': "ud_tagset/fair_by_document_type/train_nlprepl-ud.conllu.gz",
            'dev': "ud_tagset/fair_by_document_type/dev_nlprepl-ud.conllu.gz",
            'test': "ud_tagset/fair_by_document_type/test_nlprepl-ud.conllu.gz"
        }
    }
}


class NLPrePLConfig(datasets.BuilderConfig):
    """BuilderConfig for NKJP1M"""

    def __init__(self, tagset, **kwargs):
        """BuilderConfig forNKJP1M.

        Args:
          **kwargs: keyword arguments forwarded to super.
        """
        super(NLPrePLConfig, self).__init__(**kwargs)

        self.tagset = tagset


class NLPrePL(datasets.GeneratorBasedBuilder):
    """NKJP1M dataset."""

    BUILDER_CONFIGS = [
        NLPrePLConfig(
            name=t + "-" + tagset,
            version=datasets.Version("1.0.0"),
            tagset=tagset,
            description=_DESCRIPTION[t]
        )
        for t in _TYPES for tagset in _TAGSETS
    ]

    def _info(self):
        dataset, tagset = self.config.name.split("-")
        return datasets.DatasetInfo(
            description=_DESCRIPTION[dataset][tagset],
            features=datasets.Features(
                {
                    "sent_id": datasets.Value("string"),
                    "text": datasets.Value("string"),
                    "id": datasets.Value("string"),
                    "tokens": datasets.Sequence(datasets.Value("string")),
                    "lemmas": datasets.Sequence(datasets.Value("string")),
                    "upos": datasets.Sequence(datasets.Value("string")),
                    "xpos": datasets.Sequence(datasets.Value("string")),
                    "feats": datasets.Sequence(datasets.Value("string")),
                    "head": datasets.Sequence(datasets.Value("string")),
                    "deprel": datasets.Sequence(datasets.Value("string")),
                    "deps": datasets.Sequence(datasets.Value("string")),
                    "misc": datasets.Sequence(datasets.Value("string")),
                }
            ),
            supervised_keys=None,
            homepage="http://nkjp.pl/",
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        dataset, tagset = self.config.name.split("-")
        urls = _URLS[dataset][tagset]
        downloaded_files = dl_manager.download_and_extract(urls)
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"filepath": downloaded_files["train"]}),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={"filepath": downloaded_files["dev"]}),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={"filepath": downloaded_files["test"]}),
        ]

    def _generate_examples(self, filepath):
        id = 0
        logger.info("⏳ Generating examples from = %s", filepath)

        with open(filepath, 'r', encoding="utf-8") as f:
            tokenlist = list(conllu.parse_incr(f))
            for sent in tokenlist:
                if "sent_id" in sent.metadata:
                    idx = sent.metadata["sent_id"]
                else:
                    idx = id

                tokens = [token["form"] for token in sent]

                if "text" in sent.metadata:
                    txt = sent.metadata["text"]
                else:
                    txt = " ".join(tokens)

                yield id, {
                    "sent_id": str(idx),
                    "text": txt,
                    "id": [token["id"] for token in sent],
                    "tokens": [token["form"] for token in sent],
                    "lemmas": [token["lemma"] for token in sent],
                    "upos": [token["upos"] for token in sent],
                    "xpos": [token["xpos"] for token in sent],
                    "feats": [str(token["feats"]) for token in sent],
                    "head": [str(token["head"]) for token in sent],
                    "deprel": [str(token["deprel"]) for token in sent],
                    "deps": [str(token["deps"]) for token in sent],
                    "misc": [str(token["misc"]) for token in sent],
                }
                id += 1