A "build to learn" implementation of the Alpha Zero algorithm written in Python that uses LightGBM (Gradient Boosted Decision Tree) in place of a Deep Neural Network for value/policy functions.

A few environments (i.e., games) are implemented: Quoridor, Connect Four, and Tic-Tac-Toe.

python3.7 play.py <environment> <species-generation> <species-generation> <time-to-move>

python3.7 play.py connect_four gbdt-1 human-1 5.0

python3.7 train_bot.py <environment> <species> <num batches>

python3.7 train_bot.py connect_four gbdt 10

cd [this directory]
pyenv install 3.7.7
pyenv virtualenv 3.7.7 alpha_my_name
pyenv local alpha_my_name
pip install -r requirements.txt

Because some wheels don't appear to be built properly, you may need to first install the following:

brew install libomp

# Try installing again
pip install -r requirements.txt