Chess AI

A chess artificial intelligence using the MTD-f algorithm for move selection and that can learn from old games. It uses the python-chess library for board representation, move generation, zobrist hashing and reading .pgn files.

Temporal difference learning, a type of reinforcement learning, was used to train the AI. The inspiration for this work was this paper.

Demo

This is a match played by the AI against itself. White has a search depth of 2, and black has a search depth of 3. Black wins.

Building

Environment setup

1. Clone the repo.
2. Install python and virtualenv.
3. Download pypy binary.
4. Run virtualenv -p <pypy binary location here> venv to set up your virtual environment.
5. Run source venv/bin/activate to activate your virtual environment.
6. Run pip install -r requirements.txt to install dependencies.
7. Run pypy main.py to play against the AI and pypy self_play.py to watch the AI play against itself.
8. Run deactivate to deactivate the virtual environment.

Note: main.py and self_play.py both have MAX_DEPTH constants that can be adjusted to change the AI’s strength. Increase it to increase the strength and decrease it to decrease the strength.

Learning from dataset

1. Find and download a database of chess games in PGN format, or use this.
2. Place the downloaded .pgn file in the project root.
3. In config.py, set GAMES_FILE_NAME to the name of the .pgn file. Feel free to tweak with other parameters too.
4. Run pypy learn.py to learn.