Go AI

Reinforcement learning on the board game Go

GymGo

This codebase depends on the OpenAI gym environment GymGo. See the documentation for installation instructions

Usage

Play against our pretrained model

Actor Critic

python play.py --boardsize=9 --model=ac --temp=0.05 --mcts=81 --render=human

Q Learning

python play.py --boardsize=9 --model=val --temp=0.01 --mcts=8 --render=human

Human rendering uses the Pyglet library to make a nice GUI for you. If you find that this doesn't work on your machine, try setting render to terminal instead

Train your own model

python3 train.py --boardsize=5

See go_ai/utils.hyperparameters() to see what other hyperparameters you can modify