Chess AI modeled after DeepMind's AlphaGo Zero.

TODO:

• Train neural network.

Dependencies:

• Python 3.6
• python-chess 0.4.1
• Keras
• TensorFlow-GPU
• CUDA 7 or higher
• cuDNN 5 or higher

Using cleaning.py and GM_training.py

• Required: GMbothall.pgn (found here: https://chess-db.com/public/downloads/gamesfordownload.jsp)
• Note: This will generate a large number of large text files, on the order of 200 GB.
• Run cleaning.py
• Run GM_training.py

Self-Play

• playground.py will generate games by having the best current model (model_live.h5) play against itself
• evaluate.py will generate games by having the best current model (model_live.h5) play against the trained model (model_train.h5)