Author: Rémi Coulom
Date: 12:34:04 06/28/03
Go up one level in this thread
On June 28, 2003 at 12:03:01, Will Singleton wrote: >On June 28, 2003 at 11:20:21, Frederic Louguet wrote: > >>I read an interesting paper from Albert Xin Jiang and Michael Buro about >>Multi-ProbCut and its implementation in Crafty. I have always been very >>skeptical about this pruning technique (for chess) but the paper is rather >>optimistic. However, the data presented does not seem very convincing from a >>statistical point of view (too few games, not enough opponents). So could Robert >>Hyatt tell us a little more about the effectiveness of this technique ? Does it >>_really_ work ? > >I don't know about crafty, but I can tell you my experience. After messing with >the values of cutoffs(c), moves(m) and reduction(r), I use c=2, m=8, and r=3/4, >depending on depth. Increasing m (number of moves to probe to try to get >cutoffs) slows you down and doesn't help accuracy very much, and any value >greater than 2 for c caused all kinds of unnecessary searching. > >I found it helps slightly in many cases, and has no effect or is slightly worse >in others. I didn't quantify it, but I didn't get quite the results the authors >did. But since it didn't seem to hurt, I left it in. > >Will I think you are confusing Multi-ProbCut with Yngvi Björnsson's Multi-Cut, which is very different. Rémi
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.