Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: round#5: game 4: junior5 beta - fritz5.03

Author: Thorsten Czub

Date: 08:31:54 08/23/98

Go up one level in this thread


On August 23, 1998 at 10:54:36, Amir Ban wrote:

>Looking at the game, I think what you see is only a result of incorrect
>evaluation. When a program misevaluates a position to think it has an advantage,
>when in fact the opponent is in control, you often see draw-by-repetition
>scores. This is because the opponent is in control of the game and often can
>force a repetition. The opponent doesn't want to force a draw, because it
>doesn't want a draw, but the program doesn't understand that, based on its wrong
>evaluation.
>
>This is behavior you can expect from all programs who overevaluate their
>position. Perhaps you associate this with Fritz because you saw many games where
>this happens to it.

I found it typical for fritz to behave this way. I don't see other program have
this behaviour so often. I would say it happens in 4 to 6  from 10 games. This
is IMO a very high %. It happens only if the opponent is as strong or stronger,
of course. If the opponent is weaker fritz will overplay it and this stuff does
not appear. But in my draw/Loss games with fritz, i have very often these
behaviour. When e.g. Markus Gille visited me we watched such a game. We found
that maybe fritz extends some check-check lines very very deep, that deep that
it believes the game is DRAW and the extentions give back a draw score and the
main-line has repetition moves... But the opponent has POSITIVE score and will
not play for a draw of course, although he maybe COULD do so...
In these situations you not only see fritz having these draw-behaviour but also
it comes not much deep. Maybe because it extends those check-lines very deep.
This was our analysis when Markus and I saw this in a game cstal-fritz and
for a long time of the game. The moves played in those DRAW-score areas of fritz
were weak moves... it seemed to be obvious that the misevaluation of the
position in those situations makes fritz play/find weak moves or no better
alternative because the search tree gets ineffective in those positions. Maybe
this happens in all situations fritz does not know anything. I have no idea. All
i know is that these stuff happens 4 to 6 / 10
times on my machines.

>Programs usually play terrible when their evaluation is wrong, especially when
>it's too optimistic. The reason for this is easy to understand. With Fritz my
>own experience shows that it often manages to play very effectively even when
>its evaluation is WAY OFF. How it manages to do that is a mystery.
>
>Amir

the base of the evaluation does not seem to be important. But the relative
evaluation in the tree. The base of the evaluation does maybe coe from wrong
pre-processed evaluations. But the relation of the MOVES in the tree to this
root-base may be ok...
Fritz is a very selective program... we should not forget this.



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.