This repo is a chess bot built on top of the Actor-Critic model with ClipPPO loss.
Note: The env is copied from the torchrl repo, and modified to accept san_moves.txt because the setup.py for torchrl v0.7.2 does not ship with the required moves file.
To install dependencies, run poetry env activate then poetry install
To train the model, run train.ipynb. The model params can be tuned there as well, including the save rate, and the reward function can be found under env.py. To simulate a game, run play.ipynb.