mc4_nl_cleaned.py

# coding=utf-8
# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Cleaned Dutch split of the mC4 corpus."""


import json
import gzip
import textwrap
import datasets
import random
from itertools import zip_longest

logger = datasets.logging.get_logger(__name__)

_CITATION = """
@article{JMLR:v21:20-074,
  author  = {Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu},
  title   = {Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer},
  journal = {Journal of Machine Learning Research},
  year    = {2020},
  volume  = {21},
  number  = {140},
  pages   = {1-67},
  url     = {http://jmlr.org/papers/v21/20-074.html}
}
"""

_DESCRIPTION = """\
A thoroughly cleaned version of the Dutch portion of the multilingual 
colossal, cleaned version of Common Crawl's web crawl corpus (mC4) by AllenAI.

Based on Common Crawl dataset: "https://commoncrawl.org".

This is the processed version of Google's mC4 dataset by AllenAI, with further cleaning
detailed in the repository README file.
"""

_HOMEPAGE = "https://github.com/allenai/allennlp/discussions/5056"

_LICENSE = "Open Data Commons Attribution License (ODC-By) v1.0"

_DATA_URL_NL = "https://huggingface.co/datasets/yhavinga/mc4_nl_cleaned/resolve/main/mc4_nl_cleaned/{split}/c4-nl{validation}-cleaned.tfrecord-{index:05d}-of-{n_shards:05d}.json.gz"
_DATA_URL_EN = "https://huggingface.co/datasets/allenai/c4/resolve/1ddc917116b730e1859edef32896ec5c16be51d0/{name}/c4-{split}.{index:05d}-of-{n_shards:05d}.json.gz"
_C4_EN_VARIANT = "en"

_CONFIG_NAMES = ["micro", "tiny", "small", "medium", "large", "full"]
_CONFIG_EN_NL_SUFFIX = "_en_nl"

_CONFIGS = dict(
    micro={"train": 2, "validation": 1, "estimate": "1GB"},
    tiny={"train": 100, "validation": 1, "estimate": "10GB"},
    small={"train": 250, "validation": 1, "estimate": "25GB"},
    medium={"train": 500, "validation": 2, "estimate": "50GB"},
    large={"train": 750, "validation": 3, "estimate": "75GB"},
    full={"train": 1024, "validation": 4, "estimate": "103GB"},
)


class Mc4NlCleanedConfig(datasets.BuilderConfig):
    """BuilderConfig for mC4 NL Cleaned."""

    def __init__(self, **kwargs):
        """BuilderConfig for mC4 NL Cleaned."
        Args:
            **kwargs: keyword arguments forwarded to super.
        """
        super().__init__(**kwargs)


class Mc4(datasets.GeneratorBasedBuilder):
    """mC4, a colossal, cleaned version of Common Crawl's web crawl corpus."""

    BUILDER_CONFIGS = [
        Mc4NlCleanedConfig(
            name=name,
            version=datasets.Version("1.0.0"),
            description=textwrap.dedent(
                f"""\
            A {name} cleaned version of the Dutch portion of the multilingual C4 corpus.
            Estimated size of compressed files: {_CONFIGS[name]['estimate']}
            """
            ),
        )
        for name in _CONFIG_NAMES
    ]

    BUILDER_CONFIGS += [
        Mc4NlCleanedConfig(
            name=f"{name}{_CONFIG_EN_NL_SUFFIX}",
            version=datasets.Version("1.0.0"),
            description=textwrap.dedent(
                f"""\
            A {name} cleaned version of the Dutch and English portion of the multilingual C4 corpus.
            """
            ),
        )
        for name in _CONFIG_NAMES
    ]

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "text": datasets.Value("string"),
                    "timestamp": datasets.Value("string"),
                    "url": datasets.Value("string"),
                }
            ),
            supervised_keys=None,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        data_urls = {}
        config = _CONFIGS[self.config.name.replace(_CONFIG_EN_NL_SUFFIX, "")]
        for split in ["train", "validation"]:
            start_file = config.get("start", 0) if split == "train" else 0
            num_files = config.get(split)

            data_urls[split] = []
            for index in range(start_file, start_file + num_files):
                data_urls[split].append(
                    _DATA_URL_NL.format(
                        split=split,
                        index=index,
                        validation="-validation" if split == "validation" else "",
                        n_shards=4 if split == "validation" else 1024,
                    )
                )
                if self.config.name.endswith(_CONFIG_EN_NL_SUFFIX):
                    data_urls[split].append(
                        _DATA_URL_EN.format(
                            name=_C4_EN_VARIANT,
                            split=split,
                            index=index,
                            validation="-validation" if split == "validation" else "",
                            n_shards=8 if split == "validation" else 1024,
                        )
                    )
        # Shuffle data in streaming mode, so restarts will not always start with the same data
        if dl_manager.is_streaming:
            random.shuffle(data_urls["train"])
        train_downloaded_files = dl_manager.download(data_urls["train"])
        validation_downloaded_files = dl_manager.download(data_urls["validation"])
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"filepaths": train_downloaded_files},
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={"filepaths": validation_downloaded_files},
            ),
        ]

    @staticmethod
    def grouper(iterable, n, fillvalue=None):
        """Collect data into fixed-length chunks or blocks"""
        # grouper('ABCDEFG', 3, 'x') --> ABC DEF Gxx"
        args = [iter(iterable)] * n
        return zip_longest(*args, fillvalue=fillvalue)

    @staticmethod
    def gzip_open(filepath):
        if filepath:
            return gzip.open(open(filepath, "rb"), "rt", encoding="utf-8")

    def _generate_examples(self, filepaths):
        """This function returns the examples in the raw (text) form by iterating on all the files."""
        id_ = 0
        for files in self.grouper(filepaths, 2, None):
            logger.info(f"Generating examples from {files}")
            gzip_iters = [self.gzip_open(file) for file in files if file is not None]
            for lines in zip(*gzip_iters):
                for line in lines:
                    example = json.loads(line)
                    yield id_, example
                    id_ += 1