Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: A curve fitting experiment with many chess evaluation parameters

Author: Stephen A. Boak

Date: 21:42:21 11/26/01

Go up one level in this thread


On November 26, 2001 at 13:51:48, Dann Corbit wrote:

>On November 26, 2001 at 13:30:46, Frank Phillips wrote:
>
>>Dann
>>
>>Looks interesting.  Any clue as to what it all means? :-)
>
>It will take a lot more research to sort things out.
>
>Here is the code that was used:
>ftp://cap.connx.com/pub/chess-engines/new-approach/BEOWULF.ZIP
>
>You just run the "calibrate" command.
>
>The routines that you would want to peek at are in the files pers.c, utils.c and
>testdat.h
>
>The deepest searches performed so far in the experiment are only 4 plies -- and
>about 200 CPU hours were expended even at that (in total).
>
>On the other hand, up to 5 or 6 plies won't take significantly longer to run.
>But deeper tests will be far harder to perform.  I am guessing that I will only
>test single parameters are significant depths.

Unfortunately, I think, analysis by a single parameter at a time will not
readily indicate the best overall combination of parameters (used in
combination)--but it may stimulate ideas on how to best combine certain
parameters.

I contemplate that individually optimized parameters may not work well together.
 However, available time & resources naturally limit all experiments.

The experiment will produce 'results'.  The 'meaning' of those results will be
debatable, until further experiments attempt to utilize multiple parameters in
combination.

Nevertheless, I am interested in the experiment--what it is intended to gather,
and what are the possible interpretations of the results.

Good luck!
Please let us know more--thanks for this interesting and thought provoking
experiment.

--Steve



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.