Train the AI in this application to play with you. You will start with an app that doesn't know how to play, but it will learn from game to game and will finally play just as good as you do.
The AI in this app uses Q-Learning, which is a Reinforcement Learning technique. You can play yourself with the learning rate (alpha), discount factor (gamma), and epsilon values (it uses Epsilon-decreasing strategy). There are no game specific algorithms used, and the only guidelines for the agent is 100 points reward for a win and -100 points for a loss.
Have fun! I think i's great for small children who don't play Tic Tac Toe very good yet and want to learn. It can also be useful for machine learning students to see how the agent learns with different alpha, gamma, and epsilon values.
Just when you think it's too smart, you can delete it's memory and start again. But be patient, it will require a lot of games until the AI will play good!
Powered by libGDX framework.
- Train your agent. You can now make the agent play 100 games with itself. Look for the button "Train Agent" under Agent Settings.
- Bug fixes.