Author: Vincent Diepeveen
Date: 09:37:00 06/29/04
Go up one level in this thread
On June 28, 2004 at 17:33:58, Jay Scott wrote: >On June 28, 2004 at 17:25:14, Anthony Cozzie wrote: > >>Good chess is good chess. The hope is that it generates something reasonable. >>In general it has gotten similar values to the Crafty defaults. > >That's certainly an excellent start. > >>I don't think >>autotuning will ever be as good as manual tuning, but it gives you a reasonable >>starting place > >I don't think current methods are going to work as well as manual tuning, which >is based on a combination of empirical thinking and analytic thinking, because >current methods are purely empirical. But algorithms are getting smarter; math >works on them. I believe that someday automated tuning will be so much better >than manual tuning that we'll wonder how we ever got by. > >>it also tells you whether a new parameter is good or >>complete trash. > >Which is valuable in itself. Also it's a way to find bugs: "How did that passed >pawn term get a negative value?" I disagree with the assumption that it is logical that you can find out whether a parameter doesn't work. It is just that you put time in your evaluation function what is not functioning that you will then be able to improve it. Otherwise that time doesn't get invested so the program doesn't advance in that case. > Jay
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.