Sample Implementation of https://github.com/FritzFlorian/hometrainer to show how to use it for a specific game. This is not tweaked, but can be used to try out the AlphaZero algorithm on an simple example.
Clone the repo and use it as your current directory. Install the needed dependencies (tensorflow and hometrainer):
virtualenv -p python3 venv
source venv/bin/activate
pip install .
The installation might vary on your system.
The training is separated into the training_master.py and the playing_slave.py.
The first one executed and coordinates the training, the second executes selfplay games and internal/external
evaluation games. Please only run this in a trusted network as no security is configured.
The two parts of the training are separated to allow distributed execution of selfplay games on spare hardware that you got at home.
To run both:
source venv/bin/activate
python englishdraughts/training_master.py
python englishdraughts/playing_slave.py
This will create an folder called work_dir containing results and statistics about the run.