Author: Tony Werten
Date: 22:25:39 02/18/04
Go up one level in this thread
On February 18, 2004 at 19:49:43, Dann Corbit wrote: >On February 18, 2004 at 19:34:29, Alex Szabo wrote: > >>Has anyone tried implementing the Deep Blue "no progress" idea? >> >>From M. Campbell, et al, AI 134(2002) p57-83: >>"Another idea in Deep Blue, implemented in both hardware and software, is a >>pruning mechanism we call "no progress". It is based on the assumption that if >>a move is good for a given side, it is best to play it earlier rather than >>later. "No progress" is implemented by detecting if the current position could >>have been reached by playing an alternate move at some earlier position on the >>search path. If so, the search is terminated with a fail low. Although this >>algorithm has only limited effect in most positions, situations which are >>somewhat blocked and have few pieces present can observe noticeable benefits." >> >>Sounds good to me. Anyone have any experience with this they can share? > >I feel kind of dense, but how is it different than any other transposition table >entry? Normally you take the transposition table value. What they suggest is when the remaining depth in the table is bigger than the remaining depth you have now, you adjust the score for getting there "slower" It's not really "no progress", it's "give a penalty ( bonus when your behind ) for fooling around rather than do the something that has to be done" Detecting real "no progress" is a lot easier and works better :) It's most usefull for positions where your engine says "Look what I nice position I have, let's keep it this way" and only shuffles its pieces around when it should take a little score drop first to reach a better position later. Tony
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.