Simple and clean implementation of Alpha Zero algorithm
Tensorflow and Keras on Python >=3.5 Tested on Keras 2.2.0 and Tensorflow 1.8.0
python3 alpha-zero.py train if you want to launch training
python3 alpha-zero.py play if you want to play against a trained policy. Make sure to have downloaded the .zip file in the release section and moved its contents inside checkpoints/ folder
In the release section you can find some checkpoints of a neural network trained to play connect four game. The training lasted about ~50 days and still it is not sufficient, because the selfplay phase is too demanding (tested with an i5-4670)
You can find more info inside tesi.pdf (italian only)
This is my thesis project for Computer Engineering bachelor's degree