Mcts improvements - Githubissues

PhilipEkfeldt commented 4 years ago

Implemented self_play function for data generation
Added test that self_play and mcts work with dummy network, will update them when final model base is merged
Various logic improvements for mcts, including storing values on nodes based on that nodes current player
Added BoardConverter interface for testing different tensor inputs for network
Updated game.result to take in player as well (needed this for mcts implementation)

Going to have a final look at this tomorrow but figured I'd create a pull request already to get feedback. Mostly need to update docstrings and add a few comments.

PhilipEkfeldt commented 4 years ago

Ready for complete review now

guidopetri commented 4 years ago

Gave you some work LOL. It looks really good though :)

PhilipEkfeldt commented 4 years ago

Gave you some work LOL. It looks really good though :)

Thanks, I'll fix the stuff tomorrow. See my comments

AlphaZeroIncubator / AlphaZero

Mcts improvements #35