Author: Dieter Buerssner
Date: 07:33:51 08/09/01
Go up one level in this thread
On August 07, 2001 at 16:03:34, Bruce Moreland wrote: >I've often heard people state that null move with R=3 is better than with R=2, >but I have never ever ever gotten a test result that indicates this. I also did quite some experiments here. Many similar to the ones you described. Like R=3 at the tips and at the root. Changing between R=2 and 3 by remaining search depth, and also dependent on the relative depth (say depth 10 wanted, and change between the R at 5). The later yields in yet another inconsistency with the hash tables... At test positions, I found no convincing difference between them all. There is one exception. I think in rook endgames mostly the higer R was better. I did test games (I think mostly 5min+3sec) vs. Crafty. And when I tried this first, almost anything, that had some sort of dynamical null move performed better. Later, when I tried to reproduce, I could not see this anymore. Now, it seems that R=2 allways is better. I think, the higer R hurts especially in positions, that look rather unforced to the engine. Say White has a pawn at f6. Black castled K-side. Now perhaps white can make 3 moves with the Q, until it reaches h6 for the mate threat. Say white sacs some material while doing this. So lines with the Q-move null will fail high for black. To a human, this can look rather simple. With high R, this can need a lot of depth. One one such oversight may of course outweigh all advantage gained by the general higher search depth. BTW. You mentioned, that Yace is using dynamical null move. It actually doesn't by default. It has the framework, and it can be changed by the user, to swithc to dynamical null move. But with the default setup, it will use R=2, and R=1 in positions with only one minor piece per side. I am using a null move algorithim, that is functionally similar to the "double null move" Vincent has explained here some times. Regards, Dieter
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.