Author: Robert Hyatt
Date: 19:59:45 08/30/99
Go up one level in this thread
On August 29, 1999 at 16:12:42, Ralf Elvsén wrote: >On August 29, 1999 at 15:44:15, Ralf Elvsén wrote: > >>On August 29, 1999 at 14:20:16, Mark Young wrote: >> >>>On August 29, 1999 at 13:30:20, Robert Hyatt wrote: >>> >>>>On August 29, 1999 at 12:36:18, Ralf Elvsén wrote: >>>> >>>>>This is a position from a game Hiarcs 7.32 - Crafty 16.6 >>>>>Time 25 min/game, ChessBase interface etc, but that is not >>>>>important. >>>>> >>>>>6k1/1p4p1/p7/2Ppp3/1P6/P4KP1/5P1P/8 b - - 0 49 >>>>> >>>>>What is interesting is that Crafty (black) >>>>>thinks the position is about even, with a score fluctuating >>>>>around equality, while Hiarcs for several plies >>>>>has realized that it has a won position and >>>>>has a score about +2 (an exchange of rooks has just >>>>>taken place some plies ago). >>>>> >>>>>I checked 16.6 as a winboard engine, also 16.16: >>>>>same result. >>>>> >>>>>This is a serious mis-evaluation of >>>>>the pawn structure since white easily can produce >>>>>passed pawns on both sides of the board, while >>>>>blacks connected free pawns can do nothing. >>>>> >>>>>Bob, maybe you want to look into this... >>>>> >>>>> >>>>>Regards, Ralf >>>> >>>> >>>>This is a bit speculative by hiarcs... ie I have seen it do this same sort >>>>of evaluation and be dead wrong as well... >>>> >>>>the only thing I don't evaluate at present is pawn majorities which turn into >>>>distant passed pawns. It is on my todo list, and I have some code written that >>>>does pretty well but is not ready for testing on the servers yet... >>>> >>>>My version of crafty (16.17 as released) thinks black is worse here, by about >>>>1/3 of a pawn... and as it advances pawns to make passed pawns, this gets >>>>worse. But there is always danger in these positions... and I only hope that >>>>I get more right than wrong... however, +2 here? Seems hard to fathom... >>> >>>I don't know if this is a case of specuation by Hiarcs or something else. I did >>>test some other programs on this position and they tend to agree with Hiarcs >>>7.32. >>> >>>5 min search. >>> >>>Crafty 16.13 Fritz 5 Junior 5 Hiarcs 6 Hiarcs 7.32 >>> >>> +/= .43 +- 1.56 +- 2.07 +- 2.16 +- 2.23 >> >>Not being a programmer myself I can only guess what is going on. >>If Crafty gives a big bonus to blacks passed pawns and doesn't >>consider whites potential, then that is incorrect in this case. >>However, I guess a program can give the "correct" score for >>different reasons: >> >>1. A correct evaluation of the pawn structure. I can >>imagine that it is not easy to include this in the >>evaluation function. >> >>2. The obvious fact that black is one pawn up is heavily > ^^^^^ > white > >>weighted due to the fact that it is a pawn ending, where this >>usually is extremely important. This can however give >>erroneous results in other situations. >> >>Maybe one could do some testing on similar positions to >>find out how the programs "think". Unfortunately there >>are not many programmers who seem to have time to comment >>on things like this (understandably). >>Credit to Bob. >> >>Ralf many programs have a trade bonus that is tied to material. IE if you are a pawn up, you should trade pieces but not pawns... and it might be that part of this advantage is just that pieces have been traded... which would make sense in light of the other cases you have where you equalize material...
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.