Author: Tony Werten
Date: 16:00:15 12/28/02
Go up one level in this thread
On December 28, 2002 at 16:34:42, Russell Reagan wrote: >On December 28, 2002 at 15:34:38, Tony Werten wrote: > >>Almost. If the score returns alpha+1, it always means it's better. If this node >>was called itself with (alpha,alpha+1) then there is no score>alpha AND >>score<beta (hence zero window search), so no need to research. > >So this means that the range you should test for is (alpha+1, beta)? Does this >mean Bruce's code should be changed to reflect that? Something like: > >if ((val > (alpha + 1)) && (val < beta)) // ? The score you get back is either alfa or alfa+1. These values however have not much to do with the real score. Getting alfa back means <=alfa (mostly lower), getting alfa+1 means better than alfa. > >Also, what is meant by alpha+1? I have always assumed that this meant "alpha >plus the value of one pawn." From your words it sounds like alpha + 1 means >alpha + 1 regardless of whether a pawn is worth 1 or 1000. 1 is the smallest score. So if a pawn=100, 1 is 1 centipawn. > >>Well, not completely. It happens quite often that the report "better move" is >>not correct. At the root, you have to check with an open window to make sure. > >What do you mean by "open window"? (-INFINITY, INFINITY)? Mostly a non zero window, non nescessarily a "full" window. I normally open the window gradually. So when a root move fails high, I test again with (oldscore,20) to give the move a change to "prove itself". I it fails high again I start trusting the move a little and open up the window to 1.5 pawns. The idea is that that's enough most of the time, but saves me time compared to a full open window. Tony > >Russell
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.