search

Zeta36 / chess-alpha-zero

Chess reinforcement learning by AlphaGo Zero methods.
MIT License
2.13k stars 479 forks source link

Updated best model after more supervised learning #70

Closed brianprichardson closed 6 years ago

brianprichardson commented 6 years ago

256x7 NN supervised training with several million games. Should be about 700 Elo better than before; my guess is around 2,000 Elo, but feel free to try and/or test.