src/update_descriptive_statistics.py

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

Example use:

    python update_descriptive_statistics.py --dataset wikisource

"""

import argparse
import json
import logging
import multiprocessing
from dataclasses import dataclass
from pathlib import Path
from textwrap import dedent
from typing import Self, cast

import pandas as pd
import plotnine as pn
from datasets import Dataset, load_dataset
from transformers import AutoTokenizer

from git_utilities import check_is_ancestor, get_current_revision, get_latest_revision
from tests.readme_parsing import get_tag_content, read_frontmatter_and_body, replace_tag

logger = logging.getLogger(__name__)

repo_path = Path(__file__).parent.parent

tokenizer_name = "AI-Sweden-Models/Llama-3-8B-instruct"
tokenizer = AutoTokenizer.from_pretrained(tokenizer_name, use_fast=True)


def human_readable_large_int(value: int) -> str:
    thresholds = [
        (1_000_000_000, "B"),
        (1_000_000, "M"),
        (1_000, "K"),
    ]
    for threshold, label in thresholds:
        if value > threshold:
            return f"{value/threshold:.2f}{label}"

    return str(value)


def calculate_average_document_length(
    dataset: Dataset, text_column: str = "text"
) -> float:
    texts = sum(len(t) for t in dataset[text_column])
    return texts / len(dataset)


def _count_tokens(batch):
    return {
        "token_count": [
            len(tokens)
            for tokens in tokenizer(batch["text"], padding=False)["input_ids"]  # type: ignore
        ]
    }


def calculate_number_of_tokens(
    dataset: Dataset,
    text_column: str = "text",
) -> int:
    token_counts = dataset.map(
        _count_tokens,
        batched=True,
        batch_size=1000,
        num_proc=multiprocessing.cpu_count(),
    )
    return sum(token_counts["token_count"])


@dataclass()
class DescriptiveStatsOverview:
    number_of_samples: int
    average_document_length: float
    number_of_tokens: int
    language: str = "dan, dansk, Danish"

    @classmethod
    def from_dataset(cls, dataset: Dataset) -> Self:
        return cls(
            number_of_samples=len(dataset),
            average_document_length=calculate_average_document_length(dataset),
            number_of_tokens=calculate_number_of_tokens(dataset),
        )

    def to_markdown(self) -> str:
        format = dedent(f"""

        - **Language**: {self.language}
        - **Number of samples**: {human_readable_large_int(self.number_of_samples)}
        - **Number of tokens (Llama 3)**: {human_readable_large_int(self.number_of_tokens)}
        - **Average document length (characters)**: {self.average_document_length:.2f}

        """)
        return format

    def add_to_markdown(self, markdown: str | Path) -> str:
        return replace_tag(
            markdown=markdown, package=self.to_markdown(), tag="DESC-STATS"
        )

    def to_disk(
        self, path: Path
    ):  # TODO: instead write this to the yaml header (and revision should not be added here)
        data = self.__dict__
        data["revision"] = get_current_revision()
        with path.with_suffix(".json").open("w") as f:
            json.dump(self.__dict__, f)

    @classmethod
    def from_disk(cls, path: Path):
        with path.open("r") as f:
            data = json.load(f)
        if "revision" in data:
            data.pop("revision")
        obj = cls(**data)
        return obj


sample_template = """
```py
{sample}
```

### Data Fields

An entry in the dataset consists of the following fields:

- `text`(`str`): The content of the document.
- `source` (`str`): The source of the document (see [Source Data](#source-data)).
- `id` (`str`): An unique identifier for each document.
- `added` (`str`): An date for when the document was added to this collection.
- `created` (`str`): An date range for when the document was originally created.
- `license` (`str`): The license of the document. The licenses vary according to the source. 
- `domain` (`str`): The domain of the source
- `metadata/source-pretty` (`str`): The long form version of the short-form source name
- `metadata/*`: Potentially additional metadata

"""


def add_sample(markdown_path: Path, dataset: Dataset, max_str_len: int = 100):
    logger.info("Adding dataset sample to readme")
    sample = dataset[0]
    for k in sample:
        if isinstance(k, str) and len(sample[k]) > max_str_len:
            sample[k] = sample[k][:max_str_len] + "[...]"

    json_sample = json.dumps(sample, indent=2, ensure_ascii=False)
    sample_str = sample_template.format(sample=json_sample)

    replace_tag(markdown=markdown_path, package=sample_str, tag="SAMPLE")


DATASET_PLOTS_template = """
<img src="./images/dist_document_length.png" width="600" style="margin-right: 10px;" />
<img>
"""


def add_descriptive_statistics_plots(
    markdown_path: Path,
    dataset: Dataset,
):
    logger.info("Adding descriptive statistics plot to readme.")
    lengths = [len(s) for s in dataset["text"]]
    df = pd.DataFrame({"lengths": lengths, "Source": dataset["source"]})

    plot = (
        pn.ggplot(df, pn.aes(x="lengths", y=pn.after_stat("count")))
        + pn.geom_histogram(bins=100)
        + pn.labs(
            x="Document Length (Characters)",
            y="Count",
            title="Distribution of Document Lengths",
        )
        + pn.theme_minimal()
        + pn.facet_wrap("Source", scales="free", ncol=3)
    )

    img_path = markdown_path.parent / "images"
    img_path.mkdir(parents=False, exist_ok=True)
    pn.ggsave(
        plot,
        img_path / "dist_document_length.png",
        dpi=500,
        width=10,
        height=10,
        units="in",
        verbose=False,
    )

    replace_tag(
        markdown=markdown_path, package=DATASET_PLOTS_template, tag="DATASET PLOTS"
    )


def add_desc_statitics(
    markdown_path: Path,
    dataset: Dataset,
    desc_stats_path: Path,
) -> None:
    logger.info("Adding descriptive statistics to readme.")
    desc_stats = DescriptiveStatsOverview.from_dataset(dataset)
    desc_stats.to_disk(desc_stats_path)
    desc_stats.add_to_markdown(markdown_path)


def update_dataset(
    dataset_path: Path,
    name: str,
    readme_name: None | str = None,
    force: bool = False,
) -> None:
    rev = get_latest_revision(dataset_path)
    desc_stats_path = dataset_path / "descriptive_stats.json"

    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 '{name}' is already up to date, skipping."
            )
            return

    readme_name = f"{name}.md" if readme_name is None else readme_name
    markdown_path = dataset_path / readme_name

    logger.info(f"Updating dataset: {name}")

    ds = load_dataset(str(repo_path), name, split="train")
    ds = cast(Dataset, ds)

    add_desc_statitics(markdown_path, ds, desc_stats_path)
    add_sample(markdown_path, ds)
    add_descriptive_statistics_plots(markdown_path, ds)


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 create_main_table(repo_path: Path = repo_path) -> tuple[pd.DataFrame, str, str]:
    frontmatter, _ = read_frontmatter_and_body(repo_path / "README.md")
    datasets = [
        cfg["config_name"]
        for cfg in frontmatter["configs"]
        if cfg["config_name"] != "default"
    ]

    table = {
        "Source": [],
        "Description": [],
        # "Domain": [], # TODO Add domain
        "N. Tokens": [],
        "License": [],
    }
    readme_references = ""
    license_references = (
        "[CC-0]: https://creativecommons.org/publicdomain/zero/1.0/legalcode.en\n"
        + "[CC-BY-SA 4.0]: https://creativecommons.org/licenses/by-sa/4.0/deed.en\n"
    )

    for dataset in datasets:
        dataset_path = repo_path / "data" / dataset
        readme_path = dataset_path / f"{dataset_path.name}.md"
        frontmatter, body = read_frontmatter_and_body(readme_path)
        desc_stats = DescriptiveStatsOverview.from_disk(
            dataset_path / "descriptive_stats.json"
        )

        short_description = get_tag_content(body, tag="SHORT DESCRIPTION").strip()[
            :-1
        ]  # to exclude "."
        license, license_name = frontmatter["license"], frontmatter["license_name"]

        table["Source"] += [f"[{dataset_path.name}]"]
        readme_references += (
            f"[{dataset_path.name}]: data/{dataset_path.name}/{dataset_path.name}.md\n"
        )

        table["License"] += [f"[{license_name}]"]
        if license == "other":
            license_references += f"[{license_name}]: ./data/{dataset_path.name}/{dataset_path.name}.md#license-information\n"
        table["Description"] += [short_description]
        table["N. Tokens"] += [desc_stats.number_of_tokens]

    # total
    table["Source"] += ["**Total**"]
    # table["Domain"] += [""]
    table["License"] += [""]
    table["Description"] += [""]
    table["N. Tokens"] += [sum(table["N. Tokens"])]

    df = pd.DataFrame.from_dict(table)
    df["N. Tokens"] = df["N. Tokens"].apply(human_readable_large_int)
    return df, readme_references, license_references


def update_main_table(repo_path: Path = repo_path) -> None:
    logger.info("Updating MAIN TABLE")
    main_table, readme_references, license_references = create_main_table(repo_path)
    readme_path = repo_path / "README.md"
    with readme_path.open("r") as f:
        markdown = f.read()
    package = f"{main_table.to_markdown(index=False)}\n\n{readme_references}\n\n{license_references}\n\n"
    markdown = replace_tag(markdown, package=package, tag="MAIN TABLE")
    with readme_path.open("w") as f:
        f.write(markdown)


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 and dataset != "default":
        dataset_path = repo_path / "data" / dataset
        update_dataset(dataset_path, dataset_path.name, force=force)
        return

    if dataset is None:
        datasets = (repo_path / "data").glob("*")
        for dataset_path in datasets:
            update_dataset(dataset_path, dataset_path.name, force=force)

    if dataset is None or dataset == "default":
        update_dataset(repo_path, "default", "README.md", force=force)
        update_main_table(repo_path)


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),
    )