Author: Robert Hyatt
Date: 11:36:50 12/07/99
Go up one level in this thread
On December 07, 1999 at 09:17:07, Vincent Diepeveen wrote: >On December 06, 1999 at 08:46:09, Robert Hyatt wrote: > >>On December 06, 1999 at 02:54:30, Harald Faber wrote: >> >>>On December 05, 1999 at 22:35:23, Michael Fuhrmann wrote: >>> >>>>Bob Hyatt stated in a recent post that all chess programs have "absolutely >>>>glaring holes." What are the specific weaknesses of some of the top programs. >>> >>>In most cases it is their own king safety. >> >> >>That is certainly a big one. > >should specify by program though. not all >gms are the same, not all programs are the same. > >>Another is over-rating the attacking chances relative to the opponent's >>king safety. IE many programs think that after Bxg3 hxg3 that white's >>king is very unsafe, when in reality, it is not with a doubled pawn at >>g3. > >Right, this problem is extreme by >the following programs/systems: > - deep thought > - deep blue > - deepviolet (at the iccserver, softwareversion PII-400 getting 22k nps) > >Not even gnuchess is having the f2g2g3 pattern detection as bad implemented >as the above 3. > >>Passed pawns. Many over-value these, to the point that they will create an >>advanced one that becomes lost laster. Others have no clue about the advantage > >Yes crafty even says 2 pawns at the third row are worth a rook >not taking into account the king. You may not be able to read a bitmap program properly. Two connected passers on the 6th are _not_ worth a rook, _unless_ the king is far enough away that it can't reach either promoting square. And neither of the pawns can be blockaded... > >Patzer seems giving a pawn at the 6th rank +3.0 or something anyway. > >Nimzo is very happy too with passed pawns. > >Basically it's tough to detect how well pawns are stopped, i bet all programs >have to some extend big problems with this as the evaluation is so dependant >upon just a few terms here. Every term is having a big impact on the evaluation >of the position. > >>of an outside (or distant) passed pawn. An ancillary point here is that even >>if a program recognizes a passed pawn as good (outside passed pawn) many don't >>recognize a 'distant majority' which is essentially the same thing. I have >>seen more than one program trade into a K+P vs K+P (multiple p's on both >>sides) ending that was dead lost, even though they might have been a pawn >>up or have a more advanced passer. > >Right, this might be a problem that will remain long time a problem. >there are many factors and reevaluations >to consider before you can apply to such a far >majority a winning bonus. > >>Some value a rook on the 7th too highly, particularly after all the pawns have >>advanced and the king is centralized. A rook on the 7th there isn't doing >>anything, yet more than one program is very happy with it. > >a problem in diep is it wants to centralize king in endgame always. >usual that's good but sometimes... > >>Some (including older versions of crafty) value connected passed pawns more >>valuable than two isolated passed pawns... This is wrong in king and pawn (only) >>endings... >> >>The list goes on and on... > >Right, however the biggest problem are closed positions. >The problem with closed positions is that the rules are sometimes the >opposite of normal positions.
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.