Author: Albert Silver
Date: 18:54:12 02/23/01
Go up one level in this thread
On February 23, 2001 at 17:07:02, John Merlino wrote:
>On February 22, 2001 at 18:53:18, John Merlino wrote:
>
>>On February 22, 2001 at 18:32:30, Albert Silver wrote:
>>
>>>I presume the numbers were conjured up by the marketing department. Of course >it would be easy to justify too: let's see, after 5 games, the program >performed at 2450 (on the P2), which means there is an error margin of... >about 200 points so we'll say it's rated "roughly" 2625 according to >tests... :-)
>>>
>>> Albert
>>
>>No conjuring at all.
>>
>>The rating is an estimate based on several things:
>>
>>1) The SSDF (an independent Swedish organization) rated the engine in the
>>previous version of Chessmaster to be approximately 2475 at tournament time
>>controls on a Pentium 90.
>>
>>2) This rating was increased by approximately 150 points based on the minimum
>>spec machine for Chessmaster 8000, which is a Pentium II-233, under the general
>>rule that every doubling of processor speed increases rating by 70. So, it was
>>assumed that a PII-233 is about 4 times faster than a P-90 (although it is
>>probably more). So, now we're at approximately 2625.
>>
>>3) The SSDF rating is FIDE, and the Chessmaster ratings are USCF, which adds
>>another 150 points (although, maybe it should only be 100 -- this is probably
>>the only suspect portion of the equation). So, now we're at approximately 2775.
>>
>>4) Several thousand games were played between the computer personalities. In
>>some of them, the opponents were actual USCF rated humans, which gave the
>>results a much more "real-world" calculation. These games resulted in a
>>fine-tuning of Chessmaster's rating, but I'm not sure what the adjustment was
>>from that.
>>Overall, the Chessmaster personality scored +63 =1 -2 against 11 humans rated
>>above 2000 (average rating of these humans was 2145). I do know that the
>>strongest player in the test was rated 2375 and he lost all 6 games against the
>>Chessmaster. USCF ratings are believed to have this rule of thumb: If a player
>>is rated 400 or more points above his opponent, he is almost guaranteed to win
>>the game. So, this further enforces the 2775 rating from item 3.
>>
>>5) A further adjustment is done based on your computer's speed.
>>
>>jm
>
>I was wrong about step 3 above. Only 100 points were added. My notes were wrong.
>So the USCF rating on a P2-233 should be about 2725, not 2775.
>
>jm
You were mistaken also on another point in 3) in that SSDF is FIDE. This is a
big confusion due to the fact that they both use chess and both use the Elo
system, but in that is a coincidence and nothing more. Neither pool interacts
with each other so that the SSDF _only_ represents the statistical information
based on programs against themselves. This certainly shows the progress of
programs against other programs from generation to generation, showing also the
effect that hardware may have, but they cannot be said to be FIDE ratings. They
have absolutely no input from results against players at any FIDE rating. Any
claim that they are the same can only be subjective. An opinion. But scientific
they can never be until they are truly crossed. IMO.
Albert
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.