A Keras implementation of Google's AlphaZero project on tic-tac-toe, connect-four and reversi. One can add other games by providing rules and hyperparameters for training.
- python >= 3.5
- numpy
- Keras with TensorFlow backend
python main.py <game> train [<timeInHours>]
python main.py <game> test [(c|h)(c|h)]
<game> can be one of tictactoe, connect4, reversi or their short forms ttt, c4, rev
<timeInHours>: training time in hours
(c|h)(c|h): c for AI, h for human, in their respective play orders. For example, ch means AI plays first and human plays second.
The human player will need to input the coordinates for each move on board, first row then column, separated by commas. For example, 0,0 indicates top-left corner, and 7,0 indicates bottom-left corner if on Reversi. For connect-four, the user will input column index.
The AI player will load model from file named _<game>.bestModel.h5 in the same directory as the script file.