Lab 10 review by Giorgio Cacopardi (s309685)

[GioC1810]

Hi Gabriele,
First of all well done, your code was very clear to me.
I find it very good that you trained the q-agent not only through a random player but also using a min max agent, this definitely allows the q table to be more accurate as more important states are learned in order to optimize the moves.
However, I have some advice that I think can make your code even better, mainly regarding your exploration-exploitation trade off balance:

• Regarding the epsilon parameter, I suggest you do a more comprehensive tuning, especially for the training phase, trying various types of decrements, from linear to exponential, so you can find the best
• About the move choice strategy on the other hand, in addition to epsilon greedy, I recommend you try additional ones such as upper bound confidence and softmax ( or boltzman) exploration, so you can see if there is one that performs better.

I hope my suggestions are helpful to you and best of luck.