Computer Chess Club Archives


Search

Terms

Messages

Subject: Re:

Author: Robert Hyatt

Date: 07:49:30 11/30/98

Go up one level in this thread


On November 30, 1998 at 04:04:08, blass uri wrote:

>
>On November 30, 1998 at 03:23:26, odell hall wrote:
>
>>Hi CCC
>>
>> I am writing this post because I percieve that there is blatant lack of respect
>>for older chess programs on the Chess computer message boards. I believe this
>>perception is completely unjustified! In fact when the whole question is
>>carefully examined I think one would find that the strength difference between
>>old and new programs are greatly exaggerated! What I am defining as old are the
>>programs that came out in say 93 and 94 which include Rebel6 Chessmaster 4000,
>>Genius3 Mchess4. Perhaps the "real" reason why ssdf does not test these older
>>programs on Pent mmx200 machines because it would show that there has actually
>>been very little improvement in chess software, thus very little reason to
>>upgrade to the newer programs. It would also show that most of us are simply
>>victims of Sales hype!
>
>you can find the expected rating of these old programs on pentium200MMX
>
>add 70 to the rating on pentium90 and you are close to the truth.
>
>Uri


I would add two things to what Uri wrote:

1.  the "rating number" is meaningless.  It is very tempting to take these
numbers and think of them in the context of FIDE ratings.  That is wrong.
For reasons given below.

2.  the difference in rating between an older program and a newer program is
often exaggerated for several reasons:  new/better book, learning, faster
hardware, relatively minor programming changes that simply fix small holes that
hurt older versions, etc.

IE you might play masterblaster version 14 vs masterblaster version 12 and
get a rating difference of 120 points, yet when you play them against a GM,
he can't tell them apart.  This isn't uncommon. In fact, it is not too difficult
to improve your performance against another computer while hurting your play
against humans.  I've done this *several* times unintentionally.

far too much importance is attached to numbers like the SSDF "ratings" or the
selective search "ratings".  Because they seem to be based on the same "scale"
as the FIDE ratings, yet they are not....  the "rating" is a statistical
estimate of outcome between any two players in the "rating pool".  if the
pools are different, the ratings are not comparable.  IE think about a group
of players that are beginners starting in one pool, and a group of GM's starting
another pool.  After a few weeks, the ratings in both pools would necessarily
be quite similar, yet what would happen if one player from one pool plays
another player from the other pool?




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.