Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: My thoughts on these positions for what it's worth.

Author: Uri Blass

Date: 05:31:16 12/23/05

Go up one level in this thread


On December 23, 2005 at 07:56:49, Peter Berger wrote:

>On December 22, 2005 at 22:54:42, Harry Schnapp wrote:
>
>>On December 22, 2005 at 21:11:01, Graham Banks wrote:
>>
>>>Before you look at my brief assesment of each position, please be aware that I'm
>>>not being critical of Harry here.
>>>Every book, professional or amateur, has some dubious lines.
>>>
>>    Shredder 9 gives FOR EACH LINE under +-1 !!!.
>>    NOT ONE EXAMPLE is proper for imbalance!!!.
>>    It's computerchess eng-eng not always Anand vs.Kramnik or
>>    Leko vs. Adams.
>>    More does not give to commentate it.
>>
>>      Harry.
>
>I disagree: here is a point where a fruitful discussion could *start*.
>
>You designed a method to create a set of starting positions for computerchess
>matches and encourage people to use the result for their tests.
>
>It is obvious that a lot of your time and effort went into creating this set of
>positions ( which I think is also the reason why you feel so emotional about
>your work). But it should still be possible to discuss the results in a rational
>way.
>
>Actually the real problem and topic that should be discussed IMHO is not in the
>resulting positions ( they can be evidence for some argument), but in the
>design.
>
>I wonder why test design is not a major topic of serious CCC discussion anyway,
>when it is about all the testing going on. People spend months if not years on
>machine and spare time ( and discussion of the results), but only very rarely is
>there a discussion of a concept before all this time is used ( and then, there
>is often only very little feedback) .
>
>Just an example: your design can be improved on *easily*. Instead of just asking
>one engine, ask two and only accept a position where both agree on being in the
>given eval range. This can be done automatically, so only takes some machine
>time.
>
>This is accepting the basic assumption that the whole idea to start with some
>engine eval is any good. I am sceptical if it is. There are counterexamples for
>both variables, opening positions that say Shredder and Fritz will evaluate with
><+-x but will be nearly won for one side, and positions outside this range that
>would result in a completely acceptable distribution of results ( especially
>ones with unusual material imbalances).
>
>I liked the Allgaeuer approach to being critical of one Nunn position mentioned
>in this thread, that started with the results of the actual games played with it
>to challenge its validity.
>
>So maybe it could be an alternate design idea to start with a huge database of
>computer(?)games and then look for opening positions that are frequent enough
>and where the results are distributed similar to the overall results. Again,
>this can be challenged easily ( and I can think of several problems with this
>approach myself :) ), but the starting point feels more natural to me.
>
>At least I think people should consider to spend more time on creating and
>discussing designs in relation to production of data - a very general point I
>guess.
>
>Peter

I agree

Note that I do not think that balanced position have to be positions when the
evaluation is close to draw and it should be positions when engines score as you
can expect from their rating in games.

Note that I also think that it may be possible to use statistics about the
positions to improve the rating list by first calculating how much position give
black or give white in rating points.

Example:If the position give 100 elo for black then
game between 2500 with white against 2600 with black should be considered as
game between 2500 with white and 2700 with black from white point of view and
game between 2400 with white and 2600 with black from black point of view.

Uri



This page took 0.03 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.