Author: Dave Gomboc
Date: 21:19:41 04/07/99
Go up one level in this thread
On April 07, 1999 at 18:03:50, Ren Wu wrote: >On April 07, 1999 at 17:09:52, David Eppstein wrote: > >>> I think the same techniques that have proved themselves in computer chess are >>> applicable to several other games, such as Shogi, Go and of course checkers >>> too. >> >>Historically alpha-beta hasn't worked for go, but it does seem to work for many >>other games. The usual explanation is go's high branching factor, but I don't >>believe that because gomoku has the same branching factor and alpha-beta works >>ok for that game. > >I've done engines for most board games except Go. Even though I'd like to write >one also, when i have time. > >In Gomoku, most move post a strong threat, and the opponent has to respond >immediately. This give you very good clue to cut the useless moves. In other >words, it is easy to control the tree size, and so reduce the brench factor. > >In Go however, most the move's value only show after many moves, or they are >strategy moves. It is very hard to evaluate a move without at least a local >search. There are ways to prune the 'bad' moves, based on human knowledge, but >i've seen none works error free, or close. In other words, there is no easy ways >to control the tree size. > >Even though most current Go program are knowledge driven, I am still interested >to see a Go program based on the techniques found in computer chess. Maybe we >will see a Chess 4.x type Go program in near future. > >Will the computer chess history repeat in computer Go? > >We will see. I think it will, once they can assess a static position well. The search takes care of the dynamics. I think that the problem in go is that programs aren't as good at knowing how good a position is. I know this is an oversimplification, please, no flames. Dave Gomboc
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.