data/memo/create.py

# /// script
# requires-python = "==3.12"
# dependencies = [
#     "datasets==3.2.0",
#     "dynaword"
# ]
# [tool.uv.sources]
# dynaword = { git = "https://huggingface.co/datasets/danish-foundation-models/danish-dynaword", rev = "00e7f2aee7f7ad2da423419f77ecbb9c0536de0d" }
# ///
"""
Script for downloading and processing the Danish Memo repository.

Note: To run this script, you need to set `GIT_LFS_SKIP_SMUDGE=1` to be able to install dynaword:

```bash
GIT_LFS_SKIP_SMUDGE=1 uv run data/memo/create.py
```

This second version fixed previous issues with the download and processing of the Danish Memo repository:
https://huggingface.co/datasets/danish-foundation-models/danish-dynaword/discussions/67
"""

import logging
import subprocess
from datetime import datetime
from pathlib import Path
from typing import Any

import pandas as pd
from datasets import Dataset

from dynaword.process_dataset import (
    add_token_count,
    ensure_column_order,
    remove_duplicate_text,
    remove_empty_texts,
)

logger = logging.getLogger(__name__)

download_path = Path(__file__).parent / "tmp"


def download_repo(
    download_path: Path = download_path,
    repo_url: str = "https://huggingface.co/datasets/MiMe-MeMo/Corpus-v1.1",
    revision="7205897f1f3ee65e296072f3e96d49488e54e8ce",
) -> Path:
    """
    Downloads the repository from the given URL to the specified path.
    """
    logger.info(f"Downloading repository to {download_path}")
    if not download_path.exists():
        download_path.mkdir(parents=True, exist_ok=True)

    repo_path = download_path / repo_url.split("/")[-1]
    if repo_path.exists():
        logger.info(f"Repository already exists at {repo_path}, skipping download.")
        return repo_path

    # Use git to clone the repository running it from the download path
    subprocess.run(["git", "clone", repo_url], check=True, cwd=download_path)
    # Checkout the specific revision
    subprocess.run(["git", "checkout", revision], check=True, cwd=repo_path)
    logger.info("Download complete.")
    return repo_path


def load_texts(repo_path: Path) -> list[dict[str, str]]:
    """
    Loads texts from the downloaded repository.
    """
    text_files_path = repo_path / "texts"
    text_files = list(text_files_path.glob("*.txt"))
    texts = []
    for file in text_files:
        name = file.stem
        with file.open("r") as f:
            content = f.read()
            texts.append({"name": name, "text": content})
    logger.info(f"Loaded {len(texts)} texts from the repository.")
    return texts


def load_memo(repo_path: Path) -> pd.DataFrame:
    texts = load_texts(repo_path)

    metadata_csv = repo_path / "MeMo-corpus-metadata-v1.1-2023-06-20.csv"
    metadata = pd.read_csv(metadata_csv)
    # remove .pdf from "filename"
    metadata["filename"] = metadata["filename"].str.replace(".pdf", "", regex=False)
    texts_df = pd.DataFrame(texts)

    text_df_fileames = set(texts_df["name"])
    metadata_filenames = set(metadata["filename"])

    text_without_metadata = [t for t in text_df_fileames if t not in metadata_filenames]

    assert (
        len(text_without_metadata) == 0
    ), f"Some texts in the repository do not have metadata: {text_without_metadata}"

    # merge texts with metadata
    merged_df = pd.merge(
        texts_df, metadata, left_on="name", right_on="filename", how="inner"
    )

    logger.info(f"Loaded {len(merged_df)} rows from the MeMo dataset.")
    return merged_df


def convert_to_dynaword_format(memo_df: pd.DataFrame) -> Dataset:
    # convert to dynaword samples
    samples: list[dict[str, Any]] = []
    for _, row in memo_df.iterrows():
        text = row["text"]
        assert isinstance(text, str), f"Text is not a string: {text}"

        # if there is a title then add it to the text
        title = row["title"] if pd.notna(row["title"]) else "Ukendt titel"
        subtitle = row["subtitle"] if pd.notna(row["subtitle"]) else ""
        title = f"{title} {subtitle}".strip()

        first_name = row["firstname"]
        last_name = row["surname"]
        pseudonym = row["pseudonym"]

        full_name = f"{first_name} {last_name}".strip()
        if not full_name:
            full_name = pseudonym if pd.notna(pseudonym) else "Ukendt forfatter"
        else:
            # add pseudonym if it exists
            if pd.notna(pseudonym) and pseudonym != full_name:
                full_name += f" (Pseudonym: {pseudonym})"

        # create a new text with the title and author
        text_new = f"{title}\n\nSkrevet af {full_name}\nPubliceret {row['year']} af {row['publisher']}\n ------- \n\n{text}"

        today = datetime.now().date()
        sample = {
            "id": row["filename"],
            "text": text_new,
            "source": "memo",
            "added": today.isoformat(),
            "created": f"{row['year']}-01-01, {row['year']}-12-31",
        }

        samples.append(sample)

    ds = Dataset.from_list(samples)
    logger.info(f"Converted to dynaword format with {len(ds)} samples.")
    return ds

    ds = convert_to_dynaword_format(memo_df)


def main():
    save_path = Path(__file__).parent / "memo.parquet"

    repo_path = download_repo(download_path)
    memo_df = load_memo(repo_path)
    ds = convert_to_dynaword_format(memo_df)

    # quality checks and processing
    ds = remove_empty_texts(ds)
    ds = remove_duplicate_text(ds)
    ds = add_token_count(ds)
    ds = ensure_column_order(ds)

    # save to parquet
    ds.to_parquet(save_path)


if __name__ == "__main__":
    log_path = Path(__file__).parent / "memo.log"
    logging.basicConfig(
        level=logging.INFO,
        format="%(asctime)s - %(levelname)s - %(message)s",
        handlers=[
            logging.StreamHandler(),
            logging.FileHandler(log_path),
        ],
    )
    main()