Author: Zappa
Date: 08:59:33 11/22/05
Go up one level in this thread
On November 22, 2005 at 11:32:53, Andrew Wagner wrote: >Two points: >1.) What I'm suggesting is quite different from tuning evaluation parameters. >What I'm trying to get at is the algorithms, structures, and heuristics >themselves. >2.) I'm quite leery of tuning an engine only against itself. Simply put, the >goal of an engine is not to beat itself, but others, so that's what it should be >trained to do. I doubt I'll have trouble finding sufficiently weak predators to >start with. > >Thanks for the input. Well, I doubt things will be so easy. I mean, suppose the difference bitboards and 0x88 is 10% in speed. That equates to say 5 elo. You need 1000 games just to measure such a change. I also read a paper on using genetic algorithms to train chess material values, and they also needed thousands of games IIRC (and that for changes with a fairly big impact). Don't listen to me, though. If everyone followed the conventional wisdom, nothing interesting would get done. anthony
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.