Author: Robert Hyatt
Date: 07:51:54 03/23/03
Go up one level in this thread
On March 23, 2003 at 01:18:11, John Boyd wrote: >Hi All, > >I was reading Ed Schroeder's page on Rebel internals and saw his comment that it >is not necessary to generate all captures in QS... only winning/equal ones... > >I quote... > >"Unlike other chess programs REBEL (since the esarly 80's) in QS does not >investigate all captures, there is absolutely no need for that, it's pretty safe >to search only the winning captures, equal captures (QxQ, RxR etc) and Queen >Promotions. Minor promotions are also excluded from QS, it's a waste of valuable >processor time. > >Excluded from the above are of course the situations when the king is in check, >all moves are generated and searched. > >For a given ply in QS REBEL will first generate and search the winning captures, >secondly when those moves do not cause a BETA cut-off then search the equal >captures and thrid and last do the checking moves (this limited to a predefined >depth, more later), the rest of the moves is skipped." > >Endquote > >Okay, so I thought I'd try this and found that it certainly made the search >faster than a scalded cat... and it blitzed the Wacnew testset but when it >played games it became much weaker than before... >I noticed also that the search was more volatile ie. it now changed its mind a >lot unless there was a definite win of material. > >Anyone else found this? Have I overlooked something? > >If this is relevant, I currently update the hashtable in the Qsearch and >generate all moves when in check, otherwise, I generate only winning/equal >captures and queen promotions as per Ed's suggestion. And there's no limit to >the Qsearch recursion... > >Any help appreciated! > >Ross many of us are doing that and have been doing it for many years. The first thing that comes to mind is how accurately are you finding "winning/even captures"? I use SEE and it is pretty accurate. If yours is not, that will cause lots of problems...
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.