[Computer-go] How many probes down the tree are necessary for a "good" bot?

Hiroshi Yamashita yss at bd.mbn.or.jp
Thu Nov 14 14:11:27 PST 2013


> When people report results like this it would be nice to know what
> version and level of Gnugo is used...

Here is current Aya's result.

playouts to get 50% against GNU Go 3.7.10 Level 10

 300 playouts on 19x19
 100 playouts on 9x9
2500 playouts, 6 free handicaps on 19x19.
 700 playouts, 6 free handicaps on 19x19. (Level 0)

Three years ago, I heard Zen 500 playouts got 50% against GNU Go
 with 4 handicaps.

Following are KGS results. Time setting is 10 minutes + 30 sec x5.

playouts  KGS rank
     600        3k
    1200        1k
    2500        1d
    5000        1d
   10000        2d  5000 threads x2
   40000        2d 10000 threads x4
   40000        3d 10000 threads x4, ponder, 1 min + 15 sec x10 Blitz

Scaling is not good. I think it is because of semeai and Life and Death

Hiroshi Yamashita

More information about the Computer-go mailing list