A simple implementation of Tic-tac-toe (board.py, game.py), and an agent learning to play via Q-learning (agent.py).
To download, open a terminal and run
https://github.com/unibe-cns/tic-tac-toe
or download the zip (top right -> Code -> Download ZIP) and unpack.
To start a game against an agent trained via self-play, open the folder of the game in a terminal and execute python main.py.
Play a game by clicking (or tapping, on touch devices) on a free field. The bot starts on the easiest level with no experience. After every time the player wins against the bot, the difficulty is increased by loading a bot that has trained for longer. If the bot wins, the game is reset.
Only tested under Linux, currently not compatible with MacOS.