AlphaFour
AlphaFour is a Connect Four AI that uses the AlphaGo Zero algorithm (nearly). The current trained version of AlphaFour is bad at ConnectFour. There are two reasons for this. Firstly, simulating games during training takes a long time, so it has only played a few hundred games against itself during training. Secondly, my implementation of Connect Four is relatively slow, which means even if the AI is given a few seconds to "think" about each move, it only has time for about two-hundred monte carlo simulations (running on my computer). This hampers the training.
Contents
connect_four.py - contains the code for the game and players
training.ipynb - train an AlphaFour player
play.ipynb - play against an AI, either a player with a simple heurstic or AlphaFour