Trainer/Engine: Add mode to autotune a profile's weights with reinforcement learning

jonthysell / Mzinga

Open-source software to play the board game Hive.

MIT License

82 stars 9 forks source link

Trainer/Engine: Add mode to autotune a profile's weights with reinforcement learning #85

Closed jonthysell closed 6 years ago

jonthysell commented 6 years ago

Basically, take the static evaluation of the board after one move, and compare it with the deep search result, and tune the weights to make them match.

See https://chessprogramming.wikispaces.com/Reinforcement%20Learning, https://chessprogramming.wikispaces.com/Meep#RootStrap

jonthysell commented 6 years ago

http://papers.nips.cc/paper/3722-bootstrapping-from-game-tree-search.pdf