Shadow is a neural network-based chess engine inspired by AlphaZero and Leela Chess. It combines the power of Monte Carlo Tree Search (MCTS) with deep learning to play strong, creative, and human-like chess.
🔥 Estimated strength: ~2500 Elo on Chess.com
🧠 Similar in spirit to AlphaZero / Leela Chess
---
- 🌲 MCTS for move selection and long-term planning
- 🧠 Neural network evaluation for position scoring and move probabilities
- 🔁 Self-play training pipeline
- ♚ Modern playstyle that mimics human intuition
- 🛠️ Custom PGN parsing, legal move masking, and optimized batch processing
- 📈 Designed for future improvements and model scaling
Chess, when played perfectly, is a draw — though this match was not perfect, it reflects the engine's strategic depth.
game.mp4
Shadow is built upon the principles of:
It simplifies the architecture while keeping core strengths of these systems — learning through self-play, using policy and value networks, and relying on search instead of brute force.
- Language: Python
- Core libraries: PyTorch, python-chess, NumPy
- Architecture: Policy + Value head neural network
- Search: Monte Carlo Tree Search (PUCT)
git clone git@github.com:Almightyoo/shadow.git
cd shadow
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt