Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Deep Blue kns compared to kns on my 3066mhz system

Author: Robert Hyatt

Date: 12:49:09 07/23/02

Go up one level in this thread


On July 23, 2002 at 15:13:14, Sune Fischer wrote:

>On July 23, 2002 at 13:53:34, Uri Blass wrote:
>
>>On July 23, 2002 at 12:46:44, Sune Fischer wrote:
>>
>>>On July 23, 2002 at 11:48:02, Uri Blass wrote:
>>>problem.
>>>>
>>>>The evaluation of the top programs of today is better relative to the evaluation
>>>>of some years ago.
>>>>
>>>>Uri
>>>
>>>I think Deep Blue had a lot of knowledge compared to the micros of 1997.
>>>
>>>It evaluated:
>>>*) concepts such as square control
>>>*) pins
>>>*) x-rays
>>>*) king safety
>>>*) pawn structure
>>>*) passed pawn
>>>*) ray control
>>>*) outposts
>>>*) pawn majority
>>>*) rook on 7th
>>>*) blockade
>>>*) restraint
>>>*) color complex
>>>*) trapped pieces
>>>*) development
>>>*) and so on...
>>>
>>>Example of kingsafety:
>>>"Before the king castles, the system computes three king safety evaluations, one
>>>for king-side castling, one for queen-side castling and the base value for
>>>staying in the center. Each of these king safety evaluations takes into account
>>>the types of pieces attacking, the soundness of the king's shelter, presence of
>>>attacking pawns, color complex around the king, and os on. The final king safety
>>>evaluation is a weighted linear combination of the three king safety
>>>evaluations."
>>>
>>>-S.
>>
>>Having more complex evaluation does not mean having
>>a better evaluation.
>>It may mean that there is more danger of bugs so you may have worse evlauation.
>
>To be honest, I think your reasoning is a bit weird ;)
>
>>I also believe that linear combination of three king safety
>>evaluations is not a good idea.
>>
>>If the king is safe only in the center then I do not care about
>>the exact king safety in the king side or the queen side because I am not going
>>to go to there.
>
>Well, that is what an evaluation is supposed to do, it's up to the search to
>choose the whether or not to do the castle based on the alternatives
>investigated in the search tree :)
>I don't think I understand your objection to this, you can't evaluate 100%
>accurate, its a sum of terms leading to an overall score, that is how
>evaluations work AFAIK. If you think eg. an open file towards your king is very
>bad, you can always give that a huge penalty.
>
>>If the king is safe only in the king side then
>>I care if I can castle there but not about average of evauations.
>
>I don't follow.
>
>>
>>Uri


I would add that "a weighted linear combination" is not "an average"...



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.