Peer review Lab 10 - Dimitri Masetta, s306130

[dimi-it]

Preface

There is a usefull readme, that combined with the well written code, although a lack of comments, enable an easy understanding of the code!

Reinforcement Learning algorithm

You implemented a Q-Learning RL that in general seems correct, but there is something that don't enable the training process to be effective, and the results prove it (agent vs random with 54 wins, 11 ties, 35 losses is almost as good as random vs random considering that the agent play always as first):

• a dynamic tuning of the parameters during the training is missing, but this is not the main problem, as their influence on the performance is low
• probably something in your "update_q_value" function don't correctly update the Qvalue, but I couldn't point to the source of the problem.

I really liked that you train your agent against different players and really admire your work to develop them but I think that you can improve your training process to greatly improve your agent performance.