Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: How accurate is the Rebel Benchmark (answer = very accurate)

Author: Jorge Pichard

Date: 07:03:27 07/20/00

Go up one level in this thread


On July 20, 2000 at 08:52:26, Chris Carson wrote:

>Ever wonder how accurate the Rebel Bench Test is.  I have.
>Well, here is some empirical data.  Nice job Ed!  :)
>
>Prog    HW   MHZ Bench ELO  Opp   +  =  - Tot
>Reb 9	Pmmx 225 2511  2599 2249  7  1  0  8
>Reb 10	P2   400 2550  2554 2268  5  2  0  7
>Reb-Cen	K6   600 2576  2537 2537  2  4  2  8
>Reb-Cen	P3   500 2557  2515 2395  4  5  1 10
>                                 18 12  3 33
>
>Looks even better if you conside that the Hoffman game was
>lost due to a hardware failure on the K6-600!  Also, the Rebel
>Bench Test is for Rebel-10, so it is no surprise that the Rebel-9
>and Rebel-Century data are a little less accurate using the Rebel-10
>data.
>
>Nice job Ed, once again, you are ahead of the curve on ratings
>predictions!  :)
>
>Best Regards,
>Chris Carson

I don't believe that testing a program with a known opponent 1 or 2 months ahead
of the match is a precise way of evaluating the strength of the program against
a human GM, since the programmer already knows what type of closed opening he
prefer, and the programmer have plenty of time to prepare against those openings
which he usually play against.

Pichard.



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.