prdev/chessbench-full-policy-value

---
license: mit
task_categories:
- tabular-classification
- other
tags:
- chess
- game-ai
- evaluation
size_categories:
- 1B<n<10B
---

# ChessBenchmate Aggregated Dataset

This dataset is a transformed version of the ChessBenchmate dataset, aggregating all legal moves and their Stockfish evaluations per chess position.

## Dataset Structure

Each record contains:
- `fen`: Chess position in FEN notation
- `moves`: Dictionary mapping UCI moves to their evaluations
- `win_prob`: Win probability from 0.0 to 1.0 (Stockfish evaluation)
- `mate`: Mate indicator (None = no forced mate, '#' = immediate checkmate, integer = mate-in-N)

## File Format

- Format: MessagePack binary (streamed records)
- Files: 1024 shards (`train-XXXXX-of-01024.msgpack`)
- Estimated: ~3.6B unique positions

## Usage

```python
import msgpack

def load_positions(filepath):
"""Stream positions from a msgpack file."""
with open(filepath, 'rb') as f:
unpacker = msgpack.Unpacker(f, raw=False)
for record in unpacker:
yield record

# Example
for record in load_positions('train-00000-of-01024.msgpack'):
fen = record['fen']
moves = record['moves']
for move, eval in moves.items():
print(f"{move}: win_prob={eval['win_prob']:.3f}, mate={eval['mate']}")
break
```

## Source

Transformed from [ChessBenchmate](https://huggingface.co/datasets/Lichess/chessbenchmate) dataset.

## License

MIT License (same as source dataset)