A super-basic implementation of Q-learning for TicTacToe in Rust, so that I improve my understanding of the lecture-material of a Reinforcement-Learning-class.

Run with cargo run -r. This first trains the quality-function-policy, and then plays a tournament against a random player. Results will be around:

Q-Learning: 98.95%
Random:     00.78%
Draws:      00.27%

I used Rust for speed and type-safety, though the code does look clunky for something as simple as TicTacToe. If rewritten, it probably should have some abstract game-traits (having states, transition-functions, rewards, etc.).