merge-upload.py

import gc
import glob
import os
import shutil
import time

import polars as pl
from huggingface_hub import HfApi

run_conversion = (
    input(
        "\nwould you like to run the CSV to parquet conversion ('n' to skip to upload)? (y/n): "
    )
    .strip()
    .lower()
)

output_dir = ""

if run_conversion == "y":
    csv_dir = input("enter directory containing CSV files: ").strip()

    while not csv_dir:
        print("error: directory cannot be empty.")
        csv_dir = input("enter directory containing CSV files: ").strip()

    os.chdir(csv_dir)
    output_dir = "parquet_dataset"

    csv_files = sorted(glob.glob("*.csv"))

    if not csv_files:
        print("no CSV files found in current directory.")
        exit(1)

    print(f"found {len(csv_files)} CSV files.")

    if os.path.exists(output_dir):
        print(f"cleaning up old '{output_dir}' directory...")
        shutil.rmtree(output_dir)

    os.makedirs(output_dir)

    chunk_size = 3
    batches = [
        csv_files[i : i + chunk_size] for i in range(0, len(csv_files), chunk_size)
    ]

    print(f"\nprocessing {len(csv_files)} files into {len(batches)} parquet files...")

    total_rows = 0
    total_cols = None

    start_time = time.time()

    for batch_idx, batch_files in enumerate(batches):
        print(
            f"\n--- processing batch {batch_idx + 1}/{len(batches)} ({len(batch_files)} files) ---"
        )

        batch_dfs = []

        for f in batch_files:
            df = pl.read_csv(
                f,
                schema_overrides={
                    "delta_start": pl.Utf8,
                    "handshake_duration": pl.Utf8,
                    "payload_bytes_skewness": pl.Utf8,
                    "payload_bytes_cov": pl.Utf8,
                    "fwd_payload_bytes_skewness": pl.Utf8,
                    "fwd_payload_bytes_cov": pl.Utf8,
                    "bwd_payload_bytes_skewness": pl.Utf8,
                    "bwd_payload_bytes_cov": pl.Utf8,
                    "fwd_skewness_header_bytes": pl.Utf8,
                    "bwd_skewness_header_bytes": pl.Utf8,
                    "packets_IAT_skewness": pl.Utf8,
                    "fwd_packets_IAT_skewness": pl.Utf8,
                    "bwd_packets_IAT_skewness": pl.Utf8,
                    "skewness_packets_delta_time": pl.Utf8,
                    "skewness_packets_delta_len": pl.Utf8,
                    "skewness_header_bytes_delta_len": pl.Utf8,
                    "skewness_payload_bytes_delta_len": pl.Utf8,
                    "cov_payload_bytes_delta_len": pl.Utf8,
                },
            )

            row_count = len(df)
            col_count = len(df.columns)
            total_rows += row_count

            print(f"  - {f}: {row_count:,} rows, {col_count} columns")

            if total_cols is None:
                total_cols = col_count
            elif col_count != total_cols:
                print(
                    f"✗ ERROR: column mismatch in {f}! expected {total_cols}, got {col_count}"
                )
                exit(1)

            batch_dfs.append(df)

        print(f"  merging batch {batch_idx + 1}...")
        combined_batch = pl.concat(batch_dfs)

        output_filename = os.path.join(output_dir, f"chunk_{batch_idx + 1:02d}.parquet")
        combined_batch.write_parquet(output_filename)
        print(f"  ✓ saved to {output_filename}")

        del batch_dfs
        del combined_batch
        del df
        gc.collect()

    end_time = time.time()
    elapsed_minutes = (end_time - start_time) / 60

    full_path = os.path.abspath(output_dir)
    print("\n" + "=" * 40)
    print(f"conversion completed in {elapsed_minutes:.2f} minutes")
    print(f"total input rows processed: {total_rows:,}")
    print(f"parquet files saved in: {full_path}")
    print("=" * 40 + "\n")

else:
    print("\nskipping conversion step...")
    output_dir = input("enter the directory path you want to upload: ").strip()
    while output_dir and not os.path.isdir(output_dir):
        print(f"✗ error: directory '{output_dir}' does not exist.")
        output_dir = input(
            "enter a valid directory path (or press Enter to cancel): "
        ).strip()

    if output_dir:
        output_dir = os.path.abspath(output_dir)

if output_dir:
    repo_id = input(
        f"\nenter your huggingface repo id to upload '{output_dir}' (or press Enter to skip upload): "
    ).strip()

    if repo_id:
        if "/" not in repo_id:
            print(
                "✗ error: invalid repo id. it must contain a forward slash '/' separating your username and repo name"
            )
        else:
            try:
                print(f"\ninitializing upload to {repo_id}...")
                api = HfApi()

                api.upload_folder(
                    folder_path=output_dir,
                    repo_id=repo_id,
                    repo_type="dataset",
                )
                print("\n✓ upload successful. dataset is now on huggingface.")
            except Exception as e:
                print(
                    "\n✗ upload failed. make sure you are logged in via 'hf auth login'."
                )
                print(f"error details: {e}")
    else:
        print(
            f"\nupload skipped. you can manually upload the folder later with:\n"
            f"`hf upload --type dataset REPO_ID {output_dir} [path_in_repo]`"
        )
else:
    print("upload skipped. you can manually upload the folder later.")