search

PacktPublishing / Deep-Reinforcement-Learning-Hands-On

Hands-on Deep Reinforcement Learning, published by Packt
MIT License
2.81k stars 1.28k forks source link

Korean chess alpha zero #63

Open boardgame1 opened 4 years ago

boardgame1 commented 4 years ago

Hello. I implemented your book's alpha zero algorithm for Korean chess. Project site is https://github.com/boardgame1/AlphaJanggi. When I train on 12000+ self play game, model is not improved, winning rate is below 50%. Could you teach me why this happen? Thanks.

boardgame1 commented 4 years ago

When I implement mini korean chess, it seems working.