Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Linear rating increase w/ linear speed increase?

Author: Chris Carson

Date: 03:04:42 01/28/00

Go up one level in this thread


On January 27, 2000 at 23:12:50, Dann Corbit wrote:

>What exactly were the two columns of information used for the calculation, and
>how were they arrived at?

Ply is the x-axis and SSDF Rating is the Y-axis.
Ply and SSDF are used for the Correlation Coef (R),
Standard Error of Measure (SEM), ect...

Note:  If Rex Chess is used for the 386/16MHZ (SSDF #1,
with rating of 1930), then, R=.9899, SEM=41.4206, Slope=255.8808,
and Y-intercept=35.9833. This also gives a sample size of 7 instead
of 6.  If you add the ELH #1 on a Pp200 (Fritz 6a, 2633), then,
R=.9913, SEM=33.0830, Slope=257.3650, Y-Int=23.8467 and sample
size will be 8 data points.  Interesting that in the last run,
the Slope (points from 6x or 1 ply increase is close to the Monty
Newborn estimate).

Note also that NPS and Ply are estimates, I could re-calibrate using
just specint95, thus no spurious data would be present.

The main problem is the small sample size.  I can accept this given
that it is only the #1 programs/Hw that we have.  The sample size can be
increased by using the top 3 to 5 program instead of the top 1, thus
giveing a sample size of 24 or 40 (but showing what the average increase
would be and not the best).  I can do that calculation if you like.  :)

I can also repeat this for the Computer vs Human results we have.
This might take a little time, so let me know which calculations
(if any) you would like.  I will be out some today attending an
afternoon seminar.  I am using EXCEL for the actual calculations.  :)

SSDF:  http://home3.swipnet.se/~w-36794/ssdf/
ELH:   http://www.elhchess.demon.co.uk/ehss.htm

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.