Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: How to Measure [knowledge based vs bean counter

Author: Uri Blass

Date: 01:16:12 11/18/00

Go up one level in this thread


On November 18, 2000 at 03:45:24, Uri Blass wrote:

>I think that one of the ways to do the evaluation more accurate is to evaluate
>more things.
>
>It is clear that there are things that it is impossible to evaluate without
>being slower and without big source code.
>
>This is the reason that I believe that it is possible to get more accurate
>evaluation with slow searchers(assuming that the slow searchers are slow not
>because of the fact that the programmer does not know to optimize the program
>for speed).
>
>Uri

I can add that the evaluation of tal was not meant to be accurate.

I remember that in one of the games of tal that thorsten played against Junior6a
evaluation of about +1.xx pawns advantage when the position was a draw.

I asked thorsten if tal is so weak in tactics and he told me that tal does not
evaluate draws as 0.00 and can evaluate positive draws with positive scores and
negative draw with negative score.

He told me that humans do the same and I believe that it is wrong.
Humans can like position if they see a forced draw for themselves but only if
they see another alternative that they have no time to calculate.
In this case the main line that they calculate is not the draw line.
Programs with their alpha beta algorithm do not do it.

I agree that the evaluation should not be objective from practical point of view
but I believe that evaluation of +1 for a draw is not logical unless you believe
that your opponent is clearly better than you.

It may be logical to give different evaluation for different draws but the
difference should be small and not +1.

Uri



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.