Author: Ingo Bauer
Date: 07:18:07 08/04/04
Go up one level in this thread
On August 03, 2004 at 14:17:02, José Carlos wrote: >On August 03, 2004 at 05:10:45, Ingo Bauer wrote: > >>Hi >> >>>http://f50.parsimony.net/forum200336/messages/2551.htm >>> >>>tried these styles... >> >>What you propose is to use different styles vs. different engines. I think this >>is pretty worthless. If someone would change the S8 settings to "fit" you >>anti-S8 setting you are lost again and we have a race of settings for every >>engine. The whole concept of comparing engines would be out of bounds and not >>practical anymore. > > I think the idea is not bad. Humans players do this all the time (I avoid >sharp tactics against a very tactical player; I exchange pieces against an >oponent who plays endgames poorly, ...). In world championships, I bet >programmers/operator make small configuration changes depending on the opponent >(and of course in the book). > Winboard tells the program the name of his opponent. I once had an idea to let >the program change some internal parameters depending on the opponent (actually, >Crafty does this when it plays a human or a computer, if I'm not mistaken). >Imagine I teach my program to: play a 100 games match against Crafty. Save the >configuration and result to a file. Do some random small changes and play >another 100 games match. Save result again... After some time, the program would >come up with a set of parameters that maximize results against Crafty. And all >automated. > Could be tremendously improved, of course, but the idea itself is not so bad. > > José C. Hi I agree with what you have written (see other posts from me in that thread) humans do it that way, the problem is if 50 people test one engine your proposed way you will get 50 results and who is comparing these results again? And what are you doing if the opponents change their style during the 100 games? You may get really bad overall setting or settings that worked once during you test-run and never again. In theory it is fine, in practice I see BIG problems. I stick with the standard settings. As you can see on this forum (or any other forum related to comp chess) it is very hard to agree on settings for a single tourney (even if the programmers agreed to a WCCC setup many people complain about it), to agree on settings for an engine (or even worse - a group of engines) is impossible! Bye Ingo
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.