push_to_hf.py

import os
import glob
import pandas as pd
from datasets import Dataset, Features, Value, Image
from tqdm import tqdm

# 🗂 Directory where your parquet files live
parquet_dir = "data/chunk-000"
hf_repo = "mjuarez4/duck_mano_v4"  # NOTE: pushing to v3 now
batch_size = 50

# Collect all parquet files
parquet_files = sorted(glob.glob(os.path.join(parquet_dir, "episode_*.parquet")))
print(f"📦 Found {len(parquet_files)} files in {parquet_dir}")

# Process in batches
for i in range(0, len(parquet_files), batch_size):
    batch_files = parquet_files[i:i + batch_size]
    dfs = []

    print(f"\n🚀 Processing batch {i // batch_size} ({len(batch_files)} files)")
    for file_path in tqdm(batch_files):
        try:
            df = pd.read_parquet(file_path)

            # Clean up all non-image columns
            for col in df.columns:
                if col not in ["observation.image.low", "observation.image.wrist"]:
                    df[col] = df[col].apply(lambda x: str(x)[:1000])

            dfs.append(df)

        except Exception as e:
            print(f"❌ Failed to process {file_path}: {e}")

    if not dfs:
        print(f"⚠️ Skipping batch {i // batch_size} — no valid data.")
        continue

    df_batch = pd.concat(dfs, ignore_index=True)

    # Define schema with both image fields
    features = Features({
        col: Image(decode=True)
        if col in ["observation.image.low", "observation.image.wrist"]
        else Value("string")
        for col in df_batch.columns
    })

    try:
        dataset = Dataset.from_pandas(df_batch, features=features)
        dataset.push_to_hub(hf_repo, split=f"batch_{i // batch_size}")
        print(f"✅ Uploaded batch {i // batch_size} to {hf_repo}")
    except Exception as e:
        print(f"❌ Push failed for batch {i // batch_size}: {e}")