Deep Pepper

MCTS-based algorithm for parallel training of a chess engine. Adapted from existing deep learning game engines such as Giraffe and AlphaZero, Deep Pepper is a clean-room implementation of a chess engine that leverages Stockfish for the opening and closing book, and learns a policy entirely through self-play.

Technologies Used

We use the following technologies to train the model and interface with the Stockfish Chess engine.

• python-chess - For handling the chess environment and gameplay.
• pytorch - For training and inference.
• Stockfish - For value function and endgame evaluation.
• Tensorboard - For visualizing training progress.

Setup Instructions

1. Run pip install -r requirements.txt to install the necessary dependencies.
2. Run python launch_script.py to start training the Chess Engine.

Acknowledgements

• Giraffe
• Alpha Zero
• StockFish