Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: SSDF Rating list

Author: Martin Schubert

Date: 23:49:18 06/13/01

Go up one level in this thread


On June 13, 2001 at 18:32:45, Bertil Eklund wrote:

>On June 13, 2001 at 17:56:08, James T. Walker wrote:
>
>>On June 13, 2001 at 16:14:33, Christophe Theron wrote:
>>
>>>On June 13, 2001 at 11:20:20, James T. Walker wrote:
>>>
>>>>On June 13, 2001 at 00:01:19, Christophe Theron wrote:
>>>>
>>>>>On June 12, 2001 at 22:50:01, James T. Walker wrote:
>>>>>
>>>>>>On June 12, 2001 at 20:54:16, stuart taylor wrote:
>>>>>>
>>>>>>>On June 12, 2001 at 18:41:58, Christophe Theron wrote:
>>>>>>>
>>>>>>>>On June 12, 2001 at 14:48:10, Thoralf Karlsson wrote:
>>>>>>>>
>>>>>>>>>  THE SSDF RATING LIST 2001-06-11   79042 games played by  219 computers
>>>>>>>>>                                           Rating   +     -  Games   Won  Oppo
>>>>>>>>>                                           ------  ---   --- -----   ---  ----
>>>>>>>>>   1 Deep Fritz  128MB K6-2 450 MHz          2653   29   -28   647   64%  2551
>>>>>>>>>   2 Gambit Tiger 2.0  128MB K6-2 450 MHz    2650   43   -40   302   67%  2528
>>>>>>>>>   3 Chess Tiger 14.0 CB 128MB K6-2 450 MHz  2632   43   -40   308   67%  2508
>>>>>>>>>   4 Fritz 6.0  128MB K6-2 450 MHz           2623   23   -23   968   64%  2520
>>>>>>>>>   5 Junior 6.0  128MB K6-2 450 MHz          2596   20   -20  1230   62%  2509
>>>>>>>>>   6 Chess Tiger 12.0 DOS 128MB K6-2 450 MHz 2576   26   -26   733   61%  2499
>>>>>>>>>   7 Fritz 5.32  128MB K6-2 450 MHz          2551   25   -25   804   58%  2496
>>>>>>>>>   8 Nimzo 7.32  128MB K6-2 450 MHz          2550   24   -23   897   58%  2491
>>>>>>>>>   9 Nimzo 8.0  128MB K6-2 450 MHz           2542   28   -28   612   54%  2511
>>>>>>>>>  10 Junior 5.0  128MB K6-2 450 MHz          2534   25   -25   790   58%  2478
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>Congratulations to Frans Morsch and Mathias Feist (and the ChessBase team).
>>>>>>>>
>>>>>>>>Deep Fritz is definitely a very tough client. You cannot lead the SSDF list by
>>>>>>>>accident, and leading it for so many years in a row is probably the best
>>>>>>>>achievement of a chess program of all times.
>>>>>>>>
>>>>>>>>If you want to sum up the history of chess programs for microcomputers, I think
>>>>>>>>you just need to remember 3 names:
>>>>>>>>* Richard Lang
>>>>>>>>* Frans Morsch and Mathias Feist
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>    Christophe
>>>>>>>
>>>>>>>The roarng absence of the name Christophe, appears of course, in the signature
>>>>>>>of the post.
>>>>>>>But I have a little question. Does Deep Fritz have any advantage in the testing
>>>>>>>e.g. the fact that it already stood at the top, long before the recent GT even
>>>>>>>arrived on the scene, and so may have had an advantageous starting point?
>>>>>>>S.Taylor
>>>>>>
>>>>>>Hello Stuart,
>>>>>>I believe that is a valid question.  I would like to know the answer.  I would
>>>>>>like to know if the SSDF "Zeros out" the book learning of say Deep Fritz before
>>>>>>starting a match with Gambit Tiger when Gambit Tiger is brand new?  I still
>>>>>>think the SSDF list is quesionable because of the differences in opponents each
>>>>>>program has to face.  I'm sure it's better than nothing but I sure wouldn't like
>>>>>>to hang my hat on a 3 point difference in SSDF ratings (or even 20 points for
>>>>>>that matter).
>>>>>>Jim
>>>>>
>>>>>
>>>>>
>>>>>I don't question the reliability of the list.
>>>>>
>>>>>It is the most reliable tool that we have to evaluate the chess programs. The
>>>>>difference in the opponents each program has to face does not matter from a
>>>>>mathematical point of view.
>>>>>
>>>>>Year after year we can see that the list is reliable. Almost all objections get
>>>>>refuted, little by little. Of course it is not absolutely perfect, but I think
>>>>>it's damn good.
>>>>>
>>>>>
>>>>>
>>>>>    Christophe
>>>>
>>>>Hello Christophe,
>>>>I think the thread got sidetracked but I disagree with your assessment of the
>>>>SSDF list.  I agree it's not perfect and it's pretty good but....   I think its
>>>>too easy to make one program come out on top by selecting the number of games
>>>>played vs certain opponents.  If you could play only one opponent and get a true
>>>>rating then there would be no problem.  We all know this is not the case.  Some
>>>>programs do better against certain opponents and worse vs others.  So if you
>>>>play more games vs the opponent you do best against it will inflate your rating.
>>>> Of course the opposite is true.  So if Program "A" plays its favorite opponent
>>>>while program "B" plays it "nemesis" more games then naturally program "A" will
>>>>look better even though they may be equal or even the opposite is true.  This
>>>>becomes very critical when the difference in rating is only a few points in
>>>>reality.  I'm not saying the SSDF does this on purpose but I'm sure they are
>>>>doing nothing to compensate for this possibility.  In my opinion the best way to
>>>>do the SSDF list would be to make all top programs play an equal number of games
>>>>against the same opponents.  That way the top programs would all play the same
>>>>number of games against the same opponents and the list would look like this:
>>>>
>>>>Name         Rating      Number of games
>>>>Program A    2600        400
>>>>Program B    2590        400
>>>>Program C    2580        400
>>>
>>>
>>>
>>>I cannot think of any real evidence that such a phenomenon exist. Can you
>>>mention amongst the top programs which program gets killed by what other
>>>program?
>>>
>>>Has someone statistical evidence of this?
>>>
>>>But anyway, even if all program meet each other, I know some people will say
>>>that there is another way to bias the results: by letting a given program to
>>>enter or not to enter the list you have an influence on the programs it is
>>>supposed to kill.
>>>
>>>It's a neverending story.
>>>
>>>
>>>
>>>    Christophe
>>
>>
>>Hello Christophe,
>>You don't have to get killed or be a killer to change the rating by a few
>>points.  The first program that comes to mind is ChessMaster.  I believe that
>>playing a "Learning" program vs a non-learning program will add rating points to
>>the learning program with more and more games played between them.  If this is
>>not the case then you could just play 500 games vs any opponent you chose and
>>your rating would be just as accurate. In any case this "bias" could be avoided
>>with a little planning.
>>Jim
>
>Ok, and what is wrong now, that favours program x or y?
>
>Bertil

I doubt that the list favours a program. But I think your idea is to play 40
games in a match, so I wonder why not play exactly 40 games. Sometimes you play
more, sometimes you play less. I don't think it's a big problem playing 39 or 42
games. But it should be no problem playing the same number. Why I would prefer
this is the statistics. The best thing for getting a good statistics for ratings
would be playing a tournament like Cadaques: every program against each other
the same number of games.

Regards, Martin



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.