Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Question for Dr. Robert Hyatt

Author: Robert Hyatt

Date: 17:00:33 12/13/99

Go up one level in this thread


On December 13, 1999 at 18:13:40, paul bedrey wrote:

>On December 13, 1999 at 16:59:00, Robert Hyatt wrote:
>
>>On December 13, 1999 at 16:23:43, John Warfield wrote:
>>
>>>On December 13, 1999 at 15:09:09, Robert Hyatt wrote:
>>>
>>>>On December 13, 1999 at 14:58:30, walter irvin wrote:
>>>>
>>>>>On December 12, 1999 at 20:40:46, Robert Hyatt wrote:
>>>>>
>>>>>>On December 12, 1999 at 17:24:52, John Warfield wrote:
>>>>>>
>>>>>>>
>>>>>>> Could you please tell me what your 2450 projection is for, on what machine?
>>>>>>>Also I am curious what you would rate the best computer programs on these
>>>>>>>platforms  1. Amd 600  2. Amd 300  3 MMX 200
>>>>>>>
>>>>>>>Thanks
>>>>>>
>>>>>>
>>>>>>I think the best of today's programs, on the best of today's hardware, is
>>>>>>playing at a 2450 (roughly) level (FIDE rating).  I don't think cutting the
>>>>>>hardware speed by a factor of 2, nor doubling the hardware speed, would make
>>>>>>any significant change...  The problem is that the programs are good tactically,
>>>>>>but have significant positional holes that speed won't help.
>>>>>>
>>>>>>I think programs will continue to improve, but only as those 'holes' are
>>>>>>covered up solidly...
>>>>>
>>>>>For what it's worth I think you are 100% correct. Depending on the program the speed of the processor plays the most important part in the current ratings increase. This is to be expected since computers rely on tactics. I would expect
>program A on a faster computer to beat the same version on a slower machine.
>
>I once had the bright idea of creating a book of my favorite anti-computer
>opening(Stonewall Attack) in a very respected commercial program. Once out of
>book I was appalled at the program's move selection. It was obvious the program
>did not understand the continuation of the attack. This leads to a question.
>Would a program benefit from an understanding of specific opening(such as king
>or queenside attack, or key stratigic squares for specific pieces) OR would this
>slow the program down too much? I know this is hypothetical but as one of the
>leading chess programmers your opinion is worth more than mine.


The answer is "yes" but the solution is very complex.  IE I recognize some
stonewall-specific things inside my eval in order to not get rolled up too
easily.  And crafty knows other things about opening theory, such as not
blocking the c-pawn in queen-pawn openings, and so forth.  But, for example,
playing the KID as black, black knows that the bishop on C8 is very crucial to
the success of the king-side attack.  However, programs (at least mine and the
ones I see play the KID as black from time-to-time) don't grasp this, and just
blow the position to useless shreds...



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.