hinglish-dump.py

# hinglish-dump.py

## About: This is a dataset script for diwank/silicone-merged
## Docs: https://huggingface.co/docs/datasets/dataset_script.html


"""Raw merged dump of Hinglish (hi-EN) datasets."""


import pandas as pd
import os

import datasets

_DESCRIPTION = """\
Raw merged dump of Hinglish (hi-EN) datasets.
"""

_HOMEPAGE = "https://huggingface.co/datasets/diwank/hinglish-dump"
_LICENSE = "MIT"

_URLS = {
    subset: f"{_HOMEPAGE}/resolve/main/data/{subset}/data.h5"
    for subset in "crowd_transliteration  hindi_romanized_dump  hindi_xlit  hinge  hinglish_norm  news2018".split() }

_FEATURE_NAMES = [
    "target_hinglish",
    "source_hindi",
    "parallel_english",
    "annotations",
    "raw_input",
    "alternates",
]

config_names = _URLS.keys()
version = datasets.Version("1.0.0")

class HinglishDumpDataset(datasets.GeneratorBasedBuilder):
    """Raw merged dump of Hinglish (hi-EN) datasets."""

    VERSION = version
    CONFIGS = config_names

    BUILDER_CONFIGS = [
        datasets.BuilderConfig(name=subset, version=version, description=f"Config for {subset}")
        for subset in config_names
    ]
    
    DEFAULT_CONFIG_NAME = None

    def _info(self):

        features = datasets.Features({
            feature: datasets.Value("string")
            for feature in _FEATURE_NAMES
        })

        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
        )

    def _split_generators(self, dl_manager):
        # dl_manager is a datasets.download.DownloadManager that can be used to download and extract URLS
        # It can accept any type or nested list/dict and will give back the same structure with the url replaced with path to local files.
        # By default the archives will be extracted and a path to a cached folder where they are extracted is returned instead of the archive

        urls = _URLS[self.config.name]
        filepath = self.data_dir = dl_manager.download_and_extract(urls)
        
        return [
            datasets.SplitGenerator(
                name=getattr(datasets.Split, "VALIDATION" if split == "eval" else split.upper()),
                gen_kwargs=dict(filepath=filepath, split=split) )
            for split in ["train", "eval", "test"]
        ]
                          
    def _generate_examples(self, filepath, split):
        df = pd.read_hdf(filepath, key=split)
        
        for i, row in enumerate(df.to_dict('records')):
            yield i, row