Author: Uri Blass
Date: 10:02:11 12/09/05
Go up one level in this thread
On December 09, 2005 at 07:14:40, Claude Le Page wrote: >Shalom Uri! >I don't look for excuses to weakness of Rybka in endgame : I just notice that in >endgame its depth is curiously low , compared to fast engines as Junior or >Shredder : these increase their depth in endgame , as Rybka decrease it >You know as well as me that Junior7 has almost no endgame knowledge , and yet it >performs very well against Hiiarcs in endgame , chiefly because Hiarcs has not >enough time left to analyze deeply >Do you find normal that , in endgame ,Rybka analyze only to depth 10 , as its >opponent analyzes up to depth 17 , and why? because it is in Zeitnot >It happens precisely against the engines that seek complications in middlegame : >Rybka performs much better with those which accept simplification >Friendly Yours >Claude Le Page I can only say that comparing depth between programs is meaningless espacially if you talk about Junior. Junior does not search deeper than other programs and has simply a different definition of depth. I also think that small depth of rybka may be result of lack of knowledge in the evaluation because if you do not know which moves to prune you need to analyze everything. Rybka seems to have bad pruning in the endgame because of not having the relevant knowledge and here is an example that composed by Eduerd Nemeth: old Fritz8 can solve it in a few seconds when Rybka cannot solve it even after many minutes at higher depth. What can be the explanation for it except lack of knowledge in Rybka about endgames? I expect knowledge based program in the endgame at least to suspect after Qxa7 Bxa7 b5 axb6 Kb7 that the pawn is unstoppable by evaluation function and note to prune by tricks like null move pruning(note that I do not claim that it is easy to write knowledge based program and note that in the relevant case of nemeth even fruit and shredder do not have the knowledge and cannot solve it in a short time and I also do not know if the trick that enable fritz8 to find it is productive in games). New game, [D]1qk4K/Qp6/pBp5/P1P5/5pp1/8/5PP1/8 b - - 0 1 Analysis by Fritz 8: 1...Qe5+! = (0.00) Depth: 7/18 00:00:00 20kN 1...Qe5+ 2.Kh7 Qh5+ 3.Kg7 Qg5+ 4.Kf8 Qf5+ = (0.00) Depth: 8/23 00:00:00 47kN 1...Qe5+ 2.Kh7 Qh5+ 3.Kg7 Qg5+ 4.Kf8 Qf5+ = (0.00) Depth: 9/25 00:00:00 85kN 1...Qe5+ 2.Kh7 Qh5+ 3.Kg7 Qg5+ 4.Kf8 Qf5+ = (0.00) Depth: 10/25 00:00:00 148kN 1...Qe5+ 2.Kh7 Qh5+ 3.Kg7 Qg5+ 4.Kf8 Qf5+ = (0.00) Depth: 11/27 00:00:00 285kN 1...Qe5+ 2.Kh7 Qh5+ 3.Kg7 Qg5+ 4.Kf8 Qf5+ = (0.00) Depth: 12/29 00:00:00 580kN 1...Qe5+ 2.Kh7 Qh5+ 3.Kg7 Qg5+ 4.Kf8 Qf5+ = (0.00) Depth: 13/31 00:00:00 1092kN 1...Qe5+ 2.Kh7 Qh5+ 3.Kg7 Qg5+ 4.Kf8 Qf5+ = (0.00) Depth: 14/33 00:00:01 2127kN 1...Qe5+ 2.Kh7 Qh5+ 3.Kg7 Qg5+ 4.Kf8 Qf5+ = (0.00) Depth: 15/37 00:00:03 4492kN 1...Qe5+ 2.Kh7 Qh5+ 3.Kg7 Qg5+ 4.Kf8 Qf5+ = (0.00) Depth: 16/39 00:00:06 10142kN 1...Qxa7! = (-0.03) Depth: 16/39 00:00:07 11606kN 1...Qxa7! = (-0.19) Depth: 16/39 00:00:08 11797kN 1...Qxa7! ³ (-0.50) Depth: 16/39 00:00:08 11983kN 1...Qxa7! µ (-1.13) Depth: 16/39 00:00:08 12203kN 1...Qxa7 2.Bxa7 b5 3.cxb6 Kb7 4.Kh7 c5 5.Kh6 c4 6.Kg5 c3 7.Kxf4 c2 8.Kxg4 c1Q 9.f4 Qd2 10.g3 Qb4 11.Kg5 -+ (-3.47) Depth: 16/39 00:00:08 12802kN 1...Qxa7 2.Bxa7 b5 3.cxb6 Kb7 4.Kh7 c5 5.Kh6 c4 6.Kg5 c3 7.Kxf4 c2 8.Kxg4 c1Q 9.f4 Qd2 10.g3 Qb4 11.Kg5 -+ (-3.47) Depth: 17/33 00:00:09 13303kN (, 09.12.2005) Uri
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.