How to train a decent model from scratch

Jump to bottom Edit New page

Michael Pang edited this page Dec 27, 2017 · 2 revisions

Since I seem to have forgotten how I did this I'll try to remember here.

Train policy on unbalanced games (zero out the weaker player) and value on material.

Current branch: https://github.com/Akababa/chess-alpha-zero/tree/nohistory