ihavnoid / leelaz-ninenine

A 9x9 fork of leela-zero, which is targeted to provide a learning exercise with limited resources
GNU General Public License v3.0
25 stars 8 forks source link

Question about training 9x9 training process #7

Open KhiemDuyNguyen opened 5 years ago

KhiemDuyNguyen commented 5 years ago

I read from this post https://github.com/leela-zero/leela-zero/issues/926?fbclid=IwAR0o-VpDhlQkuetpWFC04iA3fZBmLxanG--b6R-BhuMsGmRLgDvfyepnSqM#issuecomment-371385552 that you're able to freeze from early moves and make the network plays the end game. Then you started to unfreeze the early game to adjust...

I'm working with smaller board size 5x5 => 8x8 and want to do the same thing. What part of the code did you modify? Or how did you do so? Basically the sgf training data from starts with certain beginning moves and play from there.

Thank you very much,