Author: Martin Giepmans
Date: 09:29:59 12/30/02
Go up one level in this thread
On December 30, 2002 at 11:38:18, Vincent Diepeveen wrote: >On December 30, 2002 at 11:27:46, Martin Giepmans wrote: > >>On December 30, 2002 at 07:08:27, Ed Schröder wrote: >> >>>On December 30, 2002 at 05:53:16, Martin Giepmans wrote: >>> >>>>On December 30, 2002 at 05:30:45, Ed Schröder wrote: >>>> >>>>>As it seems I have completely mishandled explaining Reduction 2b and 2c, I have >>>>>corrected things. >>>>> >>>>>http://members.home.nl/matador/chess840.htm >>>>> >>>>>Click on the blue "update" picture, it will move you directly to the changes >>>>>made. That will be standard procedure from now on in case of errors. >>>>> >>>>>if (remaining_depth<=x && remaining_depth>1) then >>>>> { if (ALPHA < SCORE + THREAT) -> do not reduce >>>>> if (ALPHA < SCORE + THREAT + MARGIN) -> reduce depth with one ply. } >>>>> >>>>> SCORE : score of EVAL >>>>> THREAT : Queen=900, Rook=500, Bishop=300, Knight=300, Pawn=100 >>>>> MARGIN : TABLE [remaining_depth]; >>>>> >>>>> static int TABLE[]= { 00,00,10,15,20,25,25,25,25,25,25,25,25,25, >>>>> 25,25,25,25,25,25,25 ........... }; >>>>> >>>>>The idea is, if SCORE+THREAD are not going to make it to ALPHA, but with an >>>>>extra small MARGIN it will then reduce the depth. I can't remember the speed-up >>>>>this reduction gave. >>>>> >>>>>==== >>>>> >>>>>Ed >>>>> >>>>>PS, the "faulty stuff" plus Uri's correction is maybe worth a try, who knows :) >>> >>>>The < in the second formula should be > ? >>>>Otherwise the two formula's together don't make sense. >>>>(if the first is true, the second is also true) >>> >>>For free-style pseudo code it is defendable, maybe in more c-code style would >>>make it more clear? >>> >>> { if (ALPHA > SCORE + THREAT && >>> ALPHA < SCORE + THREAT + MARGIN) -> reduce depth with one ply. } >> >>Yes, this makes it more clear. >>However, I wonder if the second part of the condition (the margin part) >>is useful. Maybe it is if the current window is smaller than margin, but >>otherwise? >>Of course, the extra condition makes the tric safer, but I think it could >>also cause more search instability. >> >>BTW, this probably doesn't work in an engine that uses PVS. >>Am I right? > >Wrong. > >This doesn't work in an engine using nullmove. Unless you only test tricks. >the smaller you make margin the more you positionally are going to miss. Thank you very much, Vincent. I was asking a question about PVS, not about nullmove. Martin > >The whole pruning system of Ed has a major focus upon not missing tactics. > >That was of course worth a world title in Madrid start of the 90s. > >It is a very ingenious system. > >Nowadays we have nullmove. > >Nevertheless it gives good ideas to me to work on some forward pruning >experiments in the future. > >> >>Cheers, >>Martin >> >>> >>>ALPHA=100 >>>SCORE=90 >>>THREAT=0 >>>MARGIN=20 >>> >>>-> reduction >>> >>>ALPHA=100 >>>SCORE=60 >>>THREAT=0 >>>MARGIN=20 >>> >>>-> no reduction >>> >>>Ed >>> >>> >>>>Martin
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.