README.md

---
license: apache-2.0
task_categories:
- reinforcement-learning
language:
- en
- ru
tags:
- chess
- deep-learning
- stockfish
- pytorch
- multi-pv
- chess-engine
- reinforcement-learning
size_categories:
- 100K<n<1M
---

# Strategic Chess Dataset: Multi-PV (700K+)

Dataset containing 706,000 chess positions evaluated by Stockfish 16.1. Built for training policy-value networks (AlphaZero / Leela Chess Zero style architectures).

## Dataset Summary

* 706,000 unique board positions with Multi-PV evaluations (top 3 candidate moves per position).
* 15-channel board encoding: 12 piece channels, 1 active-color channel, 2 move-history channels.
* 5,000 targeted blunder-recovery positions extracted from reinforcement learning self-play.
* Normalized evaluations: Stockfish centipawns mapped to [-1, 1] range using tanh(cp / 300.0).

## Data Structure

Format: compressed `.npz` archive.

* `states`: float32 array of shape (N, 15, 8, 8).
* `plans`: int64 array of shape (N, 3, 1) with candidate move indices (`from_square * 64 + to_square`).
* `evals`: float32 array of shape (N,) containing normalized position scores.

## PyTorch Loading Example

```python
import numpy as np
import torch
from torch.utils.data import Dataset

class StrategicChessDataset(Dataset):
    def __init__(self, npz_path):
        data = np.load(npz_path)
        self.states = data["states"]
        self.evals = data["evals"]
        self.best_moves = data["plans"][:, 0, 0]

    def __len__(self):
        return len(self.states)

    def __getitem__(self, idx):
        state = torch.from_numpy(self.states[idx]).float()
        move = torch.tensor(self.best_moves[idx], dtype=torch.long)
        val = torch.tensor(self.evals[idx], dtype=torch.float32)
        return state, move, val
```

## License

Apache 2.0.