clean_and_deduplicate.py

import logging
from pathlib import Path
from typing import cast

from datasets import Dataset, load_dataset

from dynaword.process_dataset import (
    add_token_count,
    ensure_column_order,
    remove_duplicate_text,
    remove_empty_texts,
)
from src.tests.readme_parsing import read_frontmatter_and_body

logger = logging.getLogger(__name__)

logging.basicConfig(
    format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
    level=logging.INFO,
)

root = Path(__file__).parent

data_path = root / "data"
frontmatter, _ = read_frontmatter_and_body(root / "README.md")

for dataset_cfg in frontmatter["configs"][1:]:
    logger.info(f"Processing {dataset_cfg['config_name']}")
    _data_path = data_path / dataset_cfg["config_name"]
    readme = data_path / dataset_cfg["config_name"] / f"{dataset_cfg['config_name']}.md"
    frontmatter, _ = read_frontmatter_and_body(readme)
    ds = load_dataset(_data_path.as_posix(), split="train")

    ds = cast(Dataset, ds)

    ds = ds.remove_columns(["license", "metadata"])
    ds = add_token_count(ds)
    ds = remove_empty_texts(ds)
    ds = remove_duplicate_text(ds)
    ds = ensure_column_order(ds)

    # save dataset
    ds.to_parquet(
        _data_path / f"{dataset_cfg['config_name']}.parquet",
    )
    logger.info(f"Saved {dataset_cfg['config_name']}.parquet")
    del ds