Author: Robert Hyatt
Date: 08:26:52 06/08/98
Go up one level in this thread
On June 08, 1998 at 03:03:37, Roland Pfister wrote: >Last week we discussed what to do and not do in a nullmove subtree. >My program Patzer used to not extend in such a branch. Bob argued >against it and it sounded ok to me. > >At the weekend I tried both versions and the result is that if Patzer >doesn't extend it is a little bit faster, but not much. But in some >positions it sees the solution one iteration later! > >So I changed my program to extend even in nullmove branches. >Thanks Bob. > >Roland Here's another "crazy one" to try. In Cray Blitz, I very carefully chose where to try a null move (non-recursive, only allowing *one* in any one path). I never tried it on a "PV" node for example, since such a node can not really have a null move played. And to do so would mean that the null move search might fail high (as we hope) or it might return a score between alpha and beta, but we couldn't keep it. So I simply disallowed it. I started off doing this in Crafty, but by accident, disabled the test, and the search ran faster overall. That is I *always* try a null-move search, except for when the "null-move transposition table trick" says to not try one. And surprisingly, it was/is faster. If you aren't doing this, you should try it... and let me know if it is faster or slower. I was surprised, and it simplified the code a bit as well since I no longer have to worry about whether a node is "ALL", "CUT" or "PV" now... and that code is now gone...
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.