newsgroups.py

import datasets
import textwrap
from sklearn.datasets import fetch_20newsgroups
from sklearn.model_selection import train_test_split
import pandas as pd

_NEWSGROUPS = [
    "alt.atheism",
    "comp.graphics",
    "comp.os.ms-windows.misc",
    "comp.sys.ibm.pc.hardware",
    "comp.sys.mac.hardware",
    "comp.windows.x",
    "misc.forsale",
    "rec.autos",
    "rec.motorcycles",
    "rec.sport.baseball",
    "rec.sport.hockey",
    "sci.crypt",
    "sci.electronics",
    "sci.med",
    "sci.space",
    "soc.religion.christian",
    "talk.politics.guns",
    "talk.politics.mideast",
    "talk.politics.misc",
    "talk.religion.misc",
]

_DESCRIPTION = textwrap.dedent(
    """\
    The bydate version of the 20-newsgroup dataset fetched from scikit_learn and
    split in stratified manner into train, validation and test sets. With and
    without metadata is made available as individual config names. The test set
    from the original 20 newsgroup dataset is retained while the original train
    set is split 80:20 into train and validation sets in stratified manner based
    on the newsgroup. The 20 different newsgroup are provided as the labels
    instead of config names as specified in the official huggingface dataset.
    Newsgroups are specified as labels to provide a simplified setup for text
    classification task. The 20 different newsgroup functioning as labels are:
"""
)
_DESCRIPTION += "\n".join(f"({i+1}) {j}" for i, j in enumerate(_NEWSGROUPS))

_HOMEPAGE = "http://qwone.com/~jason/20Newsgroups/"

_CITATION = """
@inproceedings{Lang95,
    author = {Ken Lang},
    title = {Newsweeder: Learning to filter netnews}
    year = {1995}
    booktitle = {Proceedings of the Twelfth International Conference on Machine Learning}
    pages = {331-339}
    }
 """

_VERSION = datasets.utils.Version("2.0.0")


class NewsgroupsConfig(datasets.BuilderConfig):
    def __init__(self, **kwargs):
        super(NewsgroupsConfig, self).__init__(version=_VERSION, **kwargs)


class Newsgroups(datasets.GeneratorBasedBuilder):

    BUILDER_CONFIGS = [
        NewsgroupsConfig(
            name="with_metadata",
            description=textwrap.dedent(
                """\
                The original complete bydate 20-Newsgroups dataset with the headers,
                footers, and quotes metadata as intact and just the continuous
                whitespaces (including new-line) replaced by single whitespace
                characters."""
            ),
        ),
        NewsgroupsConfig(
            name="without_metadata",
            description=textwrap.dedent(
                """\
                The bydate 20-Newsgroups dataset without the headers, footers,
                and quotes metadata as well as the continuous whitespaces
                (including new-line) replaced by single whitespace characters."""
            ),
        ),
    ]

    def _info(self):

        features = datasets.Features(
            {
                "text": datasets.Value("large_string"),
                "labels": datasets.features.ClassLabel(names=_NEWSGROUPS),
            }
        )

        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            supervised_keys=None,
            homepage=_HOMEPAGE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):

        if self.config.name == "with_metadata":
            train_data = fetch_20newsgroups(subset="train", random_state=42)
            test_data = fetch_20newsgroups(subset="test", random_state=42)

        else:
            train_data = fetch_20newsgroups(
                subset="train", random_state=42, remove=("headers", "footers", "quotes")
            )
            test_data = fetch_20newsgroups(
                subset="test", random_state=42, remove=("headers", "footers", "quotes")
            )

            empty_data_idcs = set(
                [i for i, j in enumerate(train_data.data) if j.strip() == ""]
            )
            train_data.data = [
                j for i, j in enumerate(train_data.data) if i not in empty_data_idcs
            ]
            train_data.target = [
                j for i, j in enumerate(train_data.target) if i not in empty_data_idcs
            ]

            empty_data_idcs = set(
                [i for i, j in enumerate(test_data.data) if j.strip() == ""]
            )
            test_data.data = [
                j for i, j in enumerate(test_data.data) if i not in empty_data_idcs
            ]
            test_data.target = [
                j for i, j in enumerate(test_data.target) if i not in empty_data_idcs
            ]

        train_labels = [train_data.target_names[i] for i in train_data.target]
        test_labels = [test_data.target_names[i] for i in test_data.target]

        train_df = pd.DataFrame({"text": train_data.data, "labels": train_labels})
        test_df = pd.DataFrame({"text": test_data.data, "labels": test_labels})

        train_df["text"] = train_df["text"].str.replace("\s+", " ", regex=True)
        test_df["text"] = test_df["text"].str.replace("\s+", " ", regex=True)

        train_df, val_df = train_test_split(
            train_df, test_size=0.2, random_state=42, stratify=train_df["labels"]
        )
        train_df = train_df.reset_index(drop=True)
        val_df = val_df.reset_index(drop=True)

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN, gen_kwargs={"df": train_df}
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION, gen_kwargs={"df": val_df}
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST, gen_kwargs={"df": test_df}
            ),
        ]

    def _generate_examples(self, df):
        for idx, row in df.iterrows():
            yield idx, row.to_dict()