Author: Robert Hyatt
Date: 07:00:27 08/24/99
Go up one level in this thread
On August 24, 1999 at 09:02:28, Shep wrote: >On August 24, 1999 at 05:39:54, Inmann Werner wrote: > > >>It disturbed me, that in some not clear positions the search often switches, and >>that only for 1 point better (1/100 pawn) which is not real relevant, but costs >>much time cause of the research. >>So I thought about making the eval result not so perfekt (score=(score/2)*2). >>Now it switches not so often, but in "normal" positions needs more nodes (less >>cutoffs?). >> >>Is this idea dumb or worth thinking about it. >>Makes a small evaluation, where much values of different positions give the same >>value the search slow? > >While I can't comment on technical details, I wanted to note that I know at >least one top-strength program (Chess Tiger) where this situation (new move >found with eval difference 0.01) happens quite often. > >Of course it's possible that this will hurt if your evaluation is not yet at >that level of strength. > >--- >Shep This is addressable if it is seen as "bad". Whenever you back up a move to the root, rather than storing that score, you can store that score + N, where N is some constant. Then, in order for another move to replace this move/score, it will have to be better than this score by at least N. If you do all your ordering/evaluation at the root, and have a very simple leaf evaluator, this sort of trick would let you put more faith in your heavy root evaluation. I've never done this myself, however...
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.