Author: Uri Blass
Date: 11:33:47 01/04/02
Go up one level in this thread
On January 04, 2002 at 12:05:17, Dieter Buerssner wrote: >On January 04, 2002 at 10:47:51, Uri Blass wrote: > >>> 10.01 0:27 -2.15 1.Nxb5 Rf5 2.Qb7 g3 3.gxf3 Qh6 4.Kf1 Qh1+ 5.Ke2 Qxd1+ >>>6.Kxd1 g2 7.Ke2 g1Q 8.Qe4 Qg6 9.Nd6 Rf8 10.Nxc4 Qxe4+ 11.fxe4 Bxd4 (9.714.038) >>>356.6 >>> 10.02 0:29 -2.14++ 1.Re1 fxg2 2.Kxg2 Qf5 3.Qxf5+ Rxf5 4.Ne4 Bxd4 5.Kg3 Rxf2 >>>6.Kxg4 (10.582.050) 355.3 > >>This is what shredder6(64 mbytes hash) >>did in the game that was posted >>it used almost 2 minutes on 1600 mhz and got depth 11 but could not >>avoid Nxb5 when your shredder6 could avoid Nxb5 at depth 10. >> >>note that I have not shredder6 but I suspect that >>shredder6 may have some hash table bug after seeing it. > >Such behaviour does not need to be a bug. In a different thread as an anser to >your question, GCP and myself gave some examples, why different hash table size >(and history of "prefilled entries") can make engines find moves at different >depthes. Both things can happen: less HT size makes the engine find the move >faster, as well as more HT size. I think, the more "search tricks" an engine >uses, the more often one can see such behaviour. Also it seems, that your cited >analysis was from the game, while Ingo's analysis was from the setup position. > >Regards, >Dieter Based on the game Shredder6 could not see the problem with Nxb5 even at depth 11(score of only -1.xx when based on the analaysis with no learning from history the score is -2.15 even at depth 10) Analysis from the game should be productive in most cases (otherwise the programmer can tell the program to forget the game) The fact that Shredder6's analysis in the game was worse than Shredder6's analysis when it has no learning and no history to use cause me to suspect a bug. 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.