Author: Vincent Diepeveen
Date: 07:44:53 06/28/99
Go up one level in this thread
On June 28, 1999 at 08:56:53, Inmann Werner wrote: >I use Null Move now for one year and works fine. >I make the null move and call alpha-beta with (alpha,beta,distance-3). > >Now I read, and it seems logical to me, that (beta-1,beta) would be enough. I >tried it and everything slowed down (tree bigger!)... Oops. There must be a bug in your program, or you test it at the wrong depths. To what depths did you test this (i don't wanna discuss whether it slows down at depths like 8 ply or less, cuz that's not interesting at all, considering we all get way beyond 10 ply at tournament level)? Reduction is obvously for [beta-1,beta] to [alfa,beta] >I tried to figue out why, and think now, it is because of the hash tables. >During the null move, only beta-1,beta is searched and stuffed in, and so i get >later less "finds" in hash table(the bounds say no for fails). That was, what i >figured out. Surely, this only happens in PV changing, cause there I have a more >open window. Well hashtables should actually speedup [beta-1,beta] window, as most moves and most positions are searched with [alfa,alfa+1] which most of the time is the same as [beta-1,beta] Assuming a search that uses nullwindow search >Normally, entrys in hash with null moves should have other hashpositions than if >no null move occured, and therefore the behaviour is strange. > >If I make a null move for white I change the hashadress with a special hashcode, >and if it is a black nullmove I use another code. Is this normal?, or do other >programmers do it another way? > >Werner
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.