Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: You compare apples with bananas

Author: Robert Hyatt

Date: 18:39:42 07/24/00

Go up one level in this thread


On July 24, 2000 at 18:10:20, blass uri wrote:

>On July 24, 2000 at 17:08:09, Robert Hyatt wrote:
>
>>On July 24, 2000 at 16:41:56, Chris Carson wrote:
>>
>>>On July 24, 2000 at 16:09:20, Harald Faber wrote:
>>>
>>>>DB 97 does only have 6 games and the ELO so far has nothing to do with SSDF-ELO.
>>>>When will all of you guys start to keep that in mind?
>>>>What about burning this equation in the miror:
>>>>
>>>>
>>>>SSDF-ELO <> FIDE-ELO
>>>>
>>>>
>>>>It is quite disappointing for me to see that still several people assume
>>>>SSDF-ELO = FIDE-ELO.
>>>
>>>I do not recall this statement in my post.  I did believe the relationship
>>>some time ago, but changed my opinon as more data was revealed.
>>>
>>>I think the SSDF is looking into adjustments to make the two
>>>rating systems more comparable, however, I am not sure it is
>>>really a priority for them.
>>>
>>>Best Regards,
>>>Chris Carson
>>
>>
>>An  adjustment is absolutely impossible.  The only way to "adjust" the SSDF list
>>is to take all the programs and play them in many human events, which would
>>give each program a valid FIDE rating.  But trying to take some programs and
>>compute a FIDE rating, and then using SSDF numbers to try to figure out how
>>far _all_ programs should be adjusted is simply bad statistics...
>>
>>May as well have a 100*random() tossed into _that_ formula...
>>
>>ratings between computers and computers, and between computers and humans don't
>>seem to have a lot in common...
>
>They are not the same but they have a lot of common.
>There is a positive correlation between results against humans and results
>against computers.
>
>I am sure that if you take the first 10 programs in the ssdf and the last 10
>programs in the ssdf and give both to play against humans you are going to find
>that the first 10 programs will be clearly better.
>
>There may be a difference but I believe that in almost all the cases if one
>program is clearly better against computers(difference of more than 100 elo) it
>is also better against humans.
>
>Uri

Sure... but if you take the first 10 programs in the SSDF and get them into
enough FIDE rated events, I believe that the order would change significantly.
and that is bad enough for trying to extrapolate FIDE from SSDF and vice-
versa...

There are too many degrees of freedom to do a reasonable predictor...
Hardware, good vs bad against humans, good vs bad against programs, book
traps vs computers that fail against humans...  etc...




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.