src/dynaword/update_descriptive_statistics.py

"""
A simple CLI to updates descriptive statistics on all datasets.

Example use:

    uv run  src/dynaword/update_descriptive_statistics.py --dataset wikisource

"""

import argparse
import json
import logging
from pathlib import Path
from typing import cast

import plotly.express as px
from datasets import Dataset, load_dataset

from dynaword.datasheet import DataSheet
from dynaword.descriptive_stats import DescriptiveStatsOverview
from dynaword.git_utilities import (
    check_is_ancestor,
    get_latest_revision,
)
from dynaword.paths import repo_path
from dynaword.plot_tokens_over_time import create_tokens_over_time_plot
from dynaword.tables import (
    create_grouped_table_str,
    create_overview_table,
    create_overview_table_str,
)

logger = logging.getLogger(__name__)

main_sheet = DataSheet.load_from_path(repo_path / "README.md")
_datasets = [
    cfg["config_name"]  # type: ignore
    for cfg in main_sheet.frontmatter["configs"]  # type: ignore
    if cfg["config_name"] != "default"  # type: ignore
]


logger = logging.getLogger(__name__)


def create_domain_distribution_plot(
    save_dir: Path = repo_path,
):
    df = create_overview_table(
        add_readable_tokens=False, add_total_row=False, add_readme_references=False
    )
    fig = px.sunburst(df, path=["Domain", "Source"], values="N. Tokens")

    fig.update_traces(textinfo="label+percent entry")
    fig.update_layout(title="Dataset Distribution by Domain and Source")

    img_path = save_dir / "images"
    img_path.mkdir(parents=False, exist_ok=True)
    save_path = img_path / "domain_distribution.png"
    fig.write_image(
        save_path,
        width=800,
        height=800,
        scale=2,
    )


def update_dataset(
    dataset_name: str,
    force: bool = False,
) -> None:
    dataset_path = (
        repo_path / "data" / dataset_name if dataset_name != "default" else repo_path
    )

    if dataset_name == "default":
        readme_name = "README.md"
    else:
        readme_name = f"{dataset_name}.md"

    rev = get_latest_revision(dataset_path)
    desc_stats_path = dataset_path / "descriptive_stats.json"
    markdown_path = dataset_path / readme_name

    if desc_stats_path.exists() and force is False:
        with desc_stats_path.open("r") as f:
            last_update = json.load(f).get("revision", None)

        if last_update is None:
            logger.warning(f"revision is not defined in {desc_stats_path}.")
        elif check_is_ancestor(ancestor_rev=last_update, rev=rev):
            logger.info(
                f"descriptive statistics for '{dataset_name}' is already up to date, skipping."
            )
            return

    logger.info(f"Computing descriptive stats for: {dataset_name}")
    ds = load_dataset(str(repo_path), dataset_name, split="train")
    ds = cast(Dataset, ds)
    desc_stats = DescriptiveStatsOverview.from_dataset(ds)
    desc_stats.to_disk(desc_stats_path)

    logger.info(f"Updating datasheet for: {dataset_name}")
    sheet = DataSheet.load_from_path(markdown_path)
    sheet.body = sheet.add_descriptive_stats(descriptive_stats=desc_stats)
    sheet.body = sheet.add_sample_and_description(ds)
    sheet.body = sheet.add_dataset_plots(ds, create_plot=True)

    if dataset_name == "default":
        logger.info("Updating Overview table")
        overview_table = create_overview_table_str()
        sheet.body = sheet.replace_tag(package=overview_table, tag="MAIN TABLE")
        logger.info("Updating domain table")
        domain_table = create_grouped_table_str(group="Domain")
        sheet.body = sheet.replace_tag(package=domain_table, tag="DOMAIN TABLE")
        logger.info("Updating license table")
        domain_table = create_grouped_table_str(group="License")
        sheet.body = sheet.replace_tag(package=domain_table, tag="LICENSE TABLE")
        create_domain_distribution_plot()
        create_tokens_over_time_plot()

    sheet.write_to_path()


def create_parser():
    parser = argparse.ArgumentParser(
        description="Calculated descriptive statistics of the datasets in tha data folder"
    )
    parser.add_argument(
        "--dataset",
        default=None,
        type=str,
        help="Use to specify if you only want to compute the statistics from a singular dataset.",
    )
    parser.add_argument(
        "--logging_level",
        default=20,
        type=int,
        help="Sets the logging level. Default to 20 (INFO), other reasonable levels are 10 (DEBUG) and 30 (WARNING).",
    )
    parser.add_argument(
        "--force",
        type=bool,
        default=False,
        action=argparse.BooleanOptionalAction,
        help="Should the statistics be forcefully recomputed. By default it checks the difference in commit ids.",
    )
    parser.add_argument(
        "--repo_path",
        default=str(repo_path),
        type=str,
        help="The repository where to calculate the descriptive statistics from",
    )
    return parser


def main(
    dataset: str | None = None,
    logging_level: int = 20,
    force: bool = False,
    repo_path: Path = repo_path,
) -> None:
    logging.basicConfig(level=logging_level)

    if dataset:
        update_dataset(dataset, force=force)
    else:
        for dataset_name in _datasets:
            update_dataset(dataset_name, force=force)
        update_dataset("default", force=force)


if __name__ == "__main__":
    parser = create_parser()
    args = parser.parse_args()

    main(
        args.dataset,
        logging_level=args.logging_level,
        force=args.force,
        repo_path=Path(args.repo_path),
    )