download_dataset.py

# /// script
# dependencies = [
#   "pandas",
#   "pyarrow",
#   "hf",
# ]
# ///

"""Downloader script to fetch the Nesting Tasks Dataset from Zenodo."""

import os
import urllib.request
import time


def download_file(url: str, dest_path: str) -> None:
    """Downloads a file with clean progress printouts.

    Args:
        url (str): The direct download URL.
        dest_path (str): File destination path.
    """
    print(f"📥 Downloading {os.path.basename(dest_path)}...")
    start_time = time.time()

    def reporthook(count, block_size, total_size):
        if total_size <= 0:
            return
        current_progress = count * block_size
        percent = min(100, int(current_progress * 100 / total_size))
        # Keep progress line on same terminal line
        print(
            f"\r   [{'=' * (percent // 5)}{' ' * (20 - percent // 5)}] {percent}% ({current_progress / (1024 * 1024):.1f}MB / {total_size / (1024 * 1024):.1f}MB)",
            end="",
            flush=True,
        )

    urllib.request.urlretrieve(url, dest_path, reporthook)  # nosec B310
    duration = time.time() - start_time
    print(f"\n   ✅ Completed in {duration:.1f}s!\n")


def convert_to_parquet() -> None:
    """Loads gzipped pickles with pandas and saves them as modern Parquet files.

    Cleans up the raw .gz files afterwards to keep the repository secure and light.
    """
    import pandas as pd

    files = ["tasks", "parts", "constraints", "shapes"]
    print("============================================================")
    print("🔄 Converting Pickle splits to Parquet format...")
    print("============================================================")

    for name in files:
        pickle_file = f"{name}.gz"
        parquet_file = f"{name}.parquet"

        if not os.path.exists(pickle_file):
            continue

        print(f"⚡ Processing '{pickle_file}' -> '{parquet_file}'...")
        try:
            # 1. Read pickled dataframe
            df = pd.read_pickle(pickle_file)

            # 2. Write to Parquet (removing pandas index to keep schema clean)
            df.to_parquet(parquet_file, index=False)
            print(f"   ✅ Saved {parquet_file}")

            # 3. Clean up the insecure raw pickle file
            os.remove(pickle_file)
            print(f"   🗑️ Removed raw {pickle_file}")
        except Exception as err:
            print(f"   ❌ Failed to convert {pickle_file}: {err}")
            return
    print()


def main() -> None:
    """Orchestrates the downloading of the Zenodo dataset files."""
    print("============================================================")
    print("📦 Zenodo Nesting Tasks Dataset Downloader")
    print("============================================================")

    # 1. Zenodo records API endpoints for version 1.1 of Lallier et al. (2022)
    files_to_download = {
        "tasks.gz": "https://zenodo.org/api/records/7030786/files/tasks.gz/content",
        "parts.gz": "https://zenodo.org/api/records/7030786/files/parts.gz/content",
        "constraints.gz": "https://zenodo.org/api/records/7030786/files/constraints.gz/content",
        "shapes.gz": "https://zenodo.org/api/records/7030786/files/shapes.gz/content",
    }

    # 2. Iterate and download each file directly into workspace
    for filename, url in files_to_download.items():
        # Check if either the converted parquet or the raw .gz file already exists
        parquet_name = filename.replace(".gz", ".parquet")
        if os.path.exists(parquet_name):
            print(
                f"ℹ️ File '{parquet_name}' already exists locally (converted). Skipping download.\n"
            )
        elif os.path.exists(filename):
            print(
                f"ℹ️ File '{filename}' already exists locally (raw .gz). Skipping download.\n"
            )
        else:
            try:
                download_file(url, filename)
            except Exception as err:
                print(f"❌ Failed to download {filename}: {err}")
                return

    # 3. Perform automatic conversion and cleanup
    convert_to_parquet()

    # 4. Validate and pretty-print heads of all parquet files
    print_dataset_head()

    print("============================================================")
    print("🎉 All dataset splits converted to Parquet successfully!")
    print("============================================================")
    print("💡 Next Step: To push this dataset to your Hugging Face profile, run:")
    print("   $ uv run hf upload clallier/nesting-tasks-2d . --repo-type=dataset")
    print("============================================================")


def print_dataset_head() -> None:
    """Loads and pretty-prints the first 10 rows of all Parquet files to verify conversion."""
    import pandas as pd

    # Configure pandas to show all columns without wrapping or ellipsis
    pd.set_option("display.max_columns", None)
    pd.set_option("display.width", 1000)

    files = ["tasks", "parts", "constraints", "shapes"]
    print("============================================================")
    print("🔬 Verifying Parquet Schemas (First 10 rows of each split)")
    print("============================================================")

    for name in files:
        parquet_file = f"{name}.parquet"
        if not os.path.exists(parquet_file):
            print(f"⚠️ Warning: '{parquet_file}' not found for validation.\n")
            continue

        print(f"\n📄 Split: {parquet_file}")
        print("------------------------------------------------------------")
        try:
            df = pd.read_parquet(parquet_file)
            print(df.head(10))
        except Exception as err:
            print(f"❌ Failed to read {parquet_file}: {err}")
        print("------------------------------------------------------------")
    print()


if __name__ == "__main__":
    main()