Author: Uri Blass
Date: 10:58:22 01/06/06
Go up one level in this thread
On January 06, 2006 at 12:55:27, Uri Blass wrote: >On January 06, 2006 at 12:07:11, Albert Silver wrote: > >>>>>>These numbers seem so imbalanced. I wonder if there is not some sort of bug in >>>>>>the fritz gui or something. >>>>>> >>>>>>best >>>>>>Joseph >>>>> >>>>>Your title is misleading so I changed it in my reply. >>>>>My opinion is that Rybka searches x+2 plies when it claims to search x plies. >>>>> >>>>>From my experience it can see at fixed depth of 1 ply things that other programs >>>>>see at fixed depth of 3 plies >>>>> >>>>>Here is an example >>>>> >>>>>[D]3k2r1/ppb5/3p4/r2N1p2/8/8/1P6/1K2RB1R b - - 0 1 >>>>>[D]3k2r1/pp1b4/3p4/r2N1p2/8/8/1P6/1K2RB1R b - - 0 1 >>>>> >>>>>Rybka does not capture the knight at depth 1 in the first case but capture the >>>>>knight ar depth 1 in the second case. >>>>> >>>>>Can Vasik explain why? >>>>> >>>>>evaluation? >>>>>I do not believe it. >>>>> >>>>>It must be search and rybka search after Rxd5 Bc4 black move and see that in the >>>>>first case no move save black from a loss of full rook when in the second case >>>>>it searches Rxd5 Bc4 Bc6 >>>>> >>>>>Uri >>>>Hi all, >>>> >>>>It's clearly seen that Rybka uses some form of selective search different from >>>>others' in the quescence search which is not a full width search, therefore is >>>>not counted in nps. This explains the low nps, Rybka does extra work in the QS >>>>which pays off. >>>>Obvious, that a usual 1 ply search cannot be such better even with a superb >>>>evaluation but a Rybka-type can be. >>>> >>>>Regards, >>>>László >>> >>>No >>> >>>It is not clear >>> >>>I can achieve the same effect by not counting nodes in the qsearch and writing >>>depth x when the real depth is x+2 >>> >>>It is possible that rybka does nothing special except playing better when vasik >>>tries to hide what it does by misleading information. >>> >>>Uri >> >>With all due respect, shouldn't you try to get more information before making >>such suggestions? For example, another program famous for its low full-width ply >>counts is Hiarcs, and in an exchange between Vasik and Enrico, Vasik said he >>thought that both he and Uniacke did similar things though withut seeing the >>source code, it was impossible to confirm. If Hiarcs were equally successful, >>would you also accuse Uniacke of 'misleading information'? >> >> Albert > >I showed in the past a position when rybka can search less than 10 nodes per >second on my fast hardware so it seems clear that his definition of nodes is not >correct and that rybka does a lot of search inside what is defined by it as a >node. > >I am afraid that you may even get less nodes per second in the following >position when I see no analysis in few minutes. > > >New game - Rybka 1.0 Beta 32-bit >4k3/qqqqqqqq/8/8/8/8/QQQQQQQQ/4K3 w - - 0 1 > >Analysis by Rybka 1.0 Beta 32-bit: > > >(, 06.01.2006) > > >I wonder how many nodes per second search your Rybka 1.0 Beta in that position? > >Uri It seems that this is a bad position because there is a forced mate but I think that it is easy to find a similiar position when there is no forced mate and rybka may show analysis in that case. You may try the following position [D]4kq2/qqqqqqqq/8/8/8/8/QQQQQQQQ/4K3 w - - 0 1 Rybka needs more than a minute to show analysis. This is not the best position because rybka shows 97 knodes in 2:57 minutes and I will try to find the previous position that I analyzed and improve it. 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.