Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Kramnik's 11. Rxb7 was already a CAP record!

Author: Vincent Diepeveen

Date: 10:54:34 10/13/00

Go up one level in this thread


On October 13, 2000 at 13:34:11, Dann Corbit wrote:

>On October 13, 2000 at 08:06:14, Vincent Diepeveen wrote:
>[snip]
>>with cap you can never rely on its score as it' sjust a computer search
>>score without knowing what program did the search.
>>
>>like using cap in najdorf knowing most used program is crafty is a bit
>>weird of course. First of all i tactical outsearch with a dual PIII800 in
>>a tournament game already the cap data, secondly i can't rely on it
>>at all.
>>
>>So it's not usuable. i'm not surprised that CAP captures a pawn at b7
>>at all.
>>
>>If you rely on cap you can never win a tournament in advance, so you can't
>>trust it, so you can't use it. so saying it captures a pawn here i
>>believe at once, but it doesn't change the opinion that it's a useless
>>statement as too many important moves it plays wrong.
>>
>>More as a search of DIEP will reveal.
>
>Perhaps by minimaxing the CAP data, something better can be created.  Time will
>tell.
>
>At any rate, there are many potential uses for the data.  Making a computer play
>better is one possibility, but it may also be useful for novelty search and many
>other notions.

You're still getting the point. CAP data is a heuristic which sometimes
is good sometimes is bad. Not reliable.

In my evaluation i try to not rely on that kind of heuristics too much
if i do i lose bigtime.

So is capdata. You rely on it and lose, or you don't rely on it and don't
lose. It cannot 'guide' you in any way. The only thing that can guide
you are the newest editions of grandmasters and some insight how
computers perform. Whether a computer is unhappy or happy in a position
doesn't matter, as long as it produces the right moves.

If your program is 2000 rated, then CAP looks great. if your program
is trying to achieve a bit better as that, then you can't use CAP data.

I guess this is the basic problem. You don't even get close to 2000
still with your insight, yet you're creating a book for programs which
have past that border a long time ago.

Every chessplayer who plays national competition (1800 and up in netherlands
to use a rude number, some teams are way above that not to mention
masterclass competition where this year my team didn't even qualify for)
is laughing like hell seeing the outputs of the CAP data.

Vincent





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.