Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: ISR Chess League TPR's so far!

Author: Chris Carson

Date: 05:44:32 04/17/00

Go up one level in this thread


On April 17, 2000 at 06:20:40, Jouni Uski wrote:

>On April 16, 2000 at 15:14:44, Chris Carson wrote:
>
>>Many people have been asking for this information:
>>
>>                     Opp
>>                     FIDE
>>Program HW      TPR  Avg  W D L N
>
>>Fritz 6 500 mhz 2461 2461 1 3 1 5
>
>This is quite a lot from SSDF 2721...

We do not know the SSDF for Shredder 4 or Rebel-Cen, my
guess is that they are very close to SSDF.  When you consider
the error of measure for both rating systems (error=157 for
above fritz 6 and +39, -37 for SSDF Fritz 6) then the numbers are
predictable results (95% confidence = 2 errors of measure).
I might add that human ratings fluctuate as well.  :)  Large
errors of measure for small number of games with some wins, draws,
and losses.  :)

Fritz vs Fide range (2 errors = 314) -> actual rating somewhere
between 2147 and 2775 based on 5 games.  Very high margin of
error.

SSDF range (2 errors = +78 and -74) -> actual rating somewhere
between 2647 and 2799 based on 368 games.  More reliable due to
more games.

I see little sig. stat. difference between the two ratings, however,
SSDF has said that a 25 to 50 point adjustment downward might prove
to be resonable.  Make whatever conclusions you like.  :)  Some people
expect the computers to always perform above some level.  You can say that
you are 95% conf that Frit 6 will perform at 2147 if it makes you
feel better.  :)  However, that would not be a true (or average) level
of playing strength for Fritz 6 on this hardware, but you would be
right 95% of the time (wrong 5%). :)  I prefer to base my
estimate on the more valid and reliable (more games playe) SSDF list.
Just my opinion.  :)  If you take the lower number for the computer
ratings, then you should subtract two errors of measure from the human
(FIDE) ratings to also be 95% confident of performance above that rating.
If you do that, I think the relative difference would be about the same,
based on number of games played.  :)

Best Regards,
Chris Carson



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.