Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: About ELO rating of some Byrne's (Crafty) personalities

Author: Mike Byrne

Date: 21:00:48 12/25/04

Go up one level in this thread


On December 25, 2004 at 18:00:21, Javier Gamero wrote:

>Thanks for the useful answers.
>
>I have looked in "Selective Search" ratings and Shredder 7.04 is around 2700 at
>1000MHz (it seems a bit high), so in your dual machine (2x1.7GHz) it must be
>around 2800, and then Crafty at full speed must be about 2600. It is probably a
>bit high, you did a "wild" estimation of 2550 USCF ant it could be aprox 2450
>(FIDE). So I "prefer" your 2450 value as a first approx.
>
>I have a problem with "intensity" scaling. At int=1 and at int=50 you think
>Crafty is performing 1200-1300 and 1500-1600 respectively. It translate nicely
>at the usual rule 100 points per treble speed. Being 1500-1600 similar to your
>own ELO, I'd put pretty much confidence in those values (being aware that they
>are approximations anyway).
>
>The problem is the jump from int=50 to int=10000 (full strength?). It *doesn't*
>translate nearly well to Crafty's real ELO. If int=50 is ~1550 ELO, int=10000
>would be around 2080 ELO, hundreds of point below its real full strength. This
>suppose almost 200 ELO per treble speed! (or almost 120 ELO for double speed!).
>It looks unreasonable, something is "happening" here.

Let's backup -- my numbers are estimates - if int==1 and 1200 elo, then when
int==50, Crafty must be really close to 1800/1900 on my machine and when
int=10000, then 2500/2600.

I problaby underestimated Crafty when int==50 because I generally do not not
play with mt=2 when using int and I'm "used" to playing Crafty and perhaps I do
better than I should at the lower levels.  Other people have told me that Crafty
is stronger than my estimates at the lower levels.  Your arguments supports that
theory and I tend to agree with that as well now.

>
>Why is so much difference between int=50 and int=10000? That diff is too big for
>being explained just by any slowness the new "intensity" code can produce even
>at 10000. Do you have any data about the nodes-per-sec Crafty is doing at those
>intensity values? (In my GUI, Crafty doesn't show NPS count when using
>"intensity").
>
>My ELO is similar to 1600 as well, so I cannot test usefully what level of
>strength Crafty has with int=10000, but I can try a doubly approximative
>estimation using some CM personalities :-) . But I know two chained
>approximations can lead at one coarse result.



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.