date-data.py

import os
import re
import sys
import datasets
import pandas as pd
from huggingface_hub import HfFileSystem
from typing import List

logger = datasets.logging.get_logger(name=__name__)
fs = HfFileSystem()

_CITATION = """
"""

_DESCRIPTION = """\
    This dataset contain file about datetime date. 
    It's created with purpose is practice loading dataset from hugging face hub."""

_HOMEPAGE = """\
    https://github.com/minhnv4099
"""

_REPO = "datasets/nguyenminh4099/date-data"
_BRANCH = "main"
_REPO_BRANCH = f"{_REPO}@{_BRANCH}"

_REPO_URL = f"https://huggingface.co/{_REPO}/resolve/{_BRANCH}"
_URLS = {
    'zipfile': os.path.join(_REPO_URL, "data", "{}.zip"),
    'metadata': _REPO_URL + "/metadata.parquet",
}

_CONFIGS = ['all']
_CONFIGS.extend(
    os.path.basename(file)[:-4]
    for file in fs.listdir(_REPO_BRANCH + "/data/", detail=False)
    if file.endswith('.zip')
)


# TODO: Define Dataset Builder config
class DateDataConfig(datasets.BuilderConfig):
    def __init__(
        self,
        name: str,
        **kwargs,        
    ):
        super(DateDataConfig, self).__init__(
            name=name,
            version=datasets.Version("1.0.0"),
        )
        # self.metadata = metadata
        # self.url = kwargs.get('url', "https://huggingface.co/datasets/nguyenminh4099/date-data")
        # self.data_url = kwargs.get('data_url', None)
        # self.description = kwargs.get('description', _DESCRIPTION)
        # logger.info('call BuilderConfig')


# TODO: Define Dataset Builder
class DateData(datasets.GeneratorBasedBuilder):
    logger.info('call dataset builder')

    BUILDER_CONFIGS = [
        DateDataConfig(
            name=name,
            # metadata=_URLS['metadata'],
            # data_url=_URLS['zipfile'].format(name),
        ) 
        for name in _CONFIGS
    ]
    DEFAULT_CONFIG_NAME = 'all'

    def _info(self) -> datasets.DatasetInfo:
        features = datasets.Features({
            "id": datasets.Value('string'),
            "dow": datasets.Value('string'),
            "month": datasets.Value('string'),
            "dom": datasets.Value('string'),
            "hour": datasets.Value('string'),
            "min": datasets.Value('string'),
            "second": datasets.Value('string'),
            "timezone": datasets.Value('string'),
            "year": datasets.Value('string'),
            "file_path": datasets.Value('string'),
        })
        print(self.config)
        return datasets.DatasetInfo(
            features=features,
            description=_DESCRIPTION,
            citation=_CITATION,
            homepage=_HOMEPAGE,
        )

    def _split_generators(
        self,
        dl_manager: datasets.DownloadManager,
    ) -> List[datasets.SplitGenerator]:
        logger.info("Call _split_generators")
    
        configs = _CONFIGS[1:5] if self.config.name == 'all' else [self.config.name]
        data_files = {
            config : _URLS['zipfile'].format(config)
            for config in configs
        }
        data_dict = dl_manager.download_and_extract(data_files)
        print(data_dict)

        return [
                datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "metadata": _URLS['metadata'],
                    "data_dict": data_dict,
                }
            )
        ]

    def _generate_examples(
        self,
        metadata: str,
        data_dict: dict,
    ) -> dict:
        logger.info("Call _generate_examples")
        infos = datasets.load_dataset(
            "parquet",
            data_files=[metadata],
            split='train',
        )
        metadata_df = infos.to_pandas()
        data_df = pd.DataFrame(
            {
                "shard" : list(data_dict.keys()),
                "data_dir" : list(data_dict.values()),
            },
            columns=['shard', 'data_dir'],
            index=range(len(data_dict))
        )

        metadata_df = metadata_df.merge(
            right=data_df,
            how='right',
            left_on='shard',
            right_on='shard',
            sort=True,
        )

        for i, sample in enumerate(metadata_df.itertuples()):
            file_name = os.path.join(
                sample.data_dir, sample.id + ".txt"
            )
            example = self._read_txt(file_name=file_name)
            example['id'] = sample.id
            example['file_path'] = file_name
            
            yield i, example

    def _read_txt(
        self,
        file_name: str,
    ) -> dict:
        with open(file=file_name, mode='r') as f:
            return self._extract_datetime(f.read())
    
    def _extract_datetime(
        self,
        datetime_string: str,
    ) -> dict:
        datetime_string = datetime_string.strip("./ ")
        components = re.split(pattern=r'[\s\:]+', string=datetime_string)

        return {
            "dow": components[0],
            "month": components[1],
            "dom": components[2],
            "hour": components[3],
            "min": components[4],
            "second": components[5],
            "timezone": components[6],
            "year": components[7],
        }
DateData()