A simple implementation of DeepMind's AlphaZero/AlphaGo Zero for Connect Four. Uses Tensorflow.
To generate training data (master process, required): python main.py
To generate training data in additional processes (optional): python main.py --assist
To train: python main.py --train
To play a game against the model: python main.py --play
After a few hours of training on a GTX 1060, it learns to play the center column of the board.