I used a youtube tutorial to implement this. I refered to his code directly in his github and I looked through other people code that are similar as well. I am not sure why it does not work in the sense that the agent does not learn :(. Bellow is my agent class.

I’m trying to write a program to do reinforcement learning for tic-tac-toe. I want the engine to recognize that if you reflect a board or rotate it you get the exact same game, so that these boards should be considered the same as each other.

I’m working on a project with my teacher, and we intended to test the ATRPO (Average Reward TRPO) against different state of the art algorithms. However, I only managed to find one library that uses it (https://github.com/sony/nnabla-rl), and sadly it’s implemented in a way where it does not support Discrete Actions, which means we couldn’t use the intended environment.

how can I get out of this problem?