Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: SSDF ratings vs Human performance rating (the data).

Author: James Robertson

Date: 18:51:01 04/12/00

Go up one level in this thread


On April 12, 2000 at 13:30:48, Chris Carson wrote:

>On April 12, 2000 at 12:55:16, James Robertson wrote:
>
>>On April 12, 2000 at 12:12:01, Chris Carson wrote:
>>
>>>					FIDE
>>>S/W			Mp    	MHZ 	TPR	SSDF
>>>Hiarcs 6		PII	350	2611	2624	450 MHZ for SSDF
>>>CM 6K			PII	450	2554	2574	200 MHZ for SSDF
>>>Fritz 6		PIII	500	2489	2721	450 MHZ for SSDF
>>
>>From your post, these are the only results on modern hardware. How can you
>>explain a 2489 TPR for Fritz, when, on slower hardware, it is rated **232**
>>points higher by the SSDF??? How about CM6k, rated 20 points higher on less than
>>one half the hardware? The only result that looks reasonable is Hiarcs', but how
>>many games has Hiarcs 6 played against humans at 40/2? Even 10 leaves a massive
>>potential error.
>>
>>Obviously, something is completely out of whack. At least for modern programs on
>>modern hardware, the list is a terrible predictor of strength against humans.
>>
>>James
>>
>>BTW, what is the difference between the SSDF results for Junior and the actual
>>results of Deep Junior in the Israeli League?
>>
>>
>>>Based on the above, the micro's had 14 performances above
>>>thier SSDF ratings against FIDE opponents and 15 performances
>>>below.  The SSDF list is very valid and reliable.
>>>
>>>Over 500 games for micro's against FIDE rated humans at 40/2.
>>>
>>>Best Regards,
>>>Chris Carson
>
>Draw whatever conclusions you like.  :)  From a stats standpoint,
>these are actual results.  :)  I do not work for any of the
>s/w companies, so I have no insight why a program performed above
>or below expected levels.  Put pressure on the companies that sell
>the programs to get more exposure with humans.  I can support that.  :)

I think that most of the older lower rated results _are_ accurate because the
list was calibrated using those results. But the results of the newer programs
whose ratings were _not_ calibrated against human lists show a remarkable
difference between their actual performances. Fritz is just the grossest example
of this, exhibiting a 232 point difference between its performance against
computers and its performance against humans.

James

>
>I should also mention Rebel-Century, no SSDF data though, but over
>20 games with FIDE rated humans on 350 to 600 MHZ machines:
>
>                        FIDE
>                        TPR
>Reb-Cen	PIII	500	2593  (includes last ISR game!)
>
>Again, sorry that the latest s/w on the latest hardware against
>the top GM's is not avaliable.  I wish it was, but this is what
>we have.  The SSDF does a great job providing us very good information.
>
>No rating is exact or static, so expect fluctuations and changes.  :)
>
>My view is that chess is chess.  :)  You may disagree.  :)
>
>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.