2021-punctuation-restoration.py

import os

import datasets
import pandas as pd


_CITATION = """"""

_DESCRIPTION = """\
This dataset is designed to be used in training models
that restore punctuation marks from the output of 
Automatic Speech Recognition system for Polish language.
"""

_HOMEPAGE = "https://github.com/poleval/2021-punctuation-restoration"

_URL = "https://huggingface.co/datasets/clarin-pl/2021-punctuation-restoration/resolve/main"

_PATHS = {
    "train": os.path.join(_URL, "train"),
    "test-A": os.path.join(_URL, "test-A"),
}


class PunctuationDatasetConfig(datasets.BuilderConfig):
    """BuilderConfig for AfrikaansNerCorpus"""

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


class PunctuationDataset(datasets.GeneratorBasedBuilder):
    """TODO: Short description of my dataset."""

    VERSION = datasets.Version("1.0.0")

    BUILDER_CONFIGS = [
        PunctuationDatasetConfig(
            name="punctuation_dataset",
            version=datasets.Version("1.0.0"),
            description="PunctuationDataset dataset",
        ),
    ]

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "text_in": datasets.Value("string"),
                    "text_out": datasets.Value("string"),
                    "tokens": datasets.Sequence(datasets.Value("string")),
                    "tags": datasets.Sequence(
                        datasets.features.ClassLabel(
                            names=[
                                'B-.',
                                'B-,',
                                'B--',
                                'B-!',
                                'B-?',
                                'B-:',
                                'B-;',
                                'O',
                            ]
                        )
                    )
                }),
            supervised_keys=None,
            homepage=_HOMEPAGE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN, gen_kwargs={"filepath": _PATHS["train"]}
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST, gen_kwargs={"filepath": _PATHS["test-A"]}
            ),
        ]

    def _generate_examples(self, filepath):
        in_df = pd.read_csv(os.path.join(filepath, "in.tsv"), sep='\t', header=None)
        out_df = pd.read_csv(os.path.join(filepath, 'expected.tsv'), sep='\t', header=None)

        for key, ((_, row_in), (_, row_out)) in enumerate(zip(in_df.iterrows(), out_df.iterrows()), 1):

            text_in = PunctuationDataset._clean_text(row_in[1])
            text_out = PunctuationDataset._clean_text(row_out[0])

            tokens = []
            tags = []
            for token_in, token_out in zip(text_in.split(), text_out.split()):
                assert token_in.lower() in token_out.lower()

                tokens.append(token_in)
                if token_in.lower() == token_out.lower():
                    tags.append('O')
                else:
                    tags.append(f'B-{token_out[-1]}')

            yield key, {
                "text_in": text_in,
                "text_out": text_out,
                "tokens": tokens,
                "tags": tags
            }

    @staticmethod
    def _clean_text(text: str, lower: bool = False) -> str:
        if lower:
            text = text.lower()
        text = text.replace(' -', '')
        text = text.replace(' .', '')
        text = text.replace(' ,', '')
        text = text.replace('  ', ' ')
        text = text.strip()
        return text