Computer Chess Club Archives




Subject: Re: evaluationfunction tuning

Author: Uri Blass

Date: 11:03:52 09/08/03

Go up one level in this thread

On September 08, 2003 at 14:00:00, Dann Corbit wrote:

>On September 07, 2003 at 16:14:54, JW de Kort wrote:
>>Dear Uri
>>Thanks for your reply. Strangly my engines plays better against Gerbil than
>>against TSCP. My search is not to bad and far more complicated than TSCP's. My
>>better search has no effect due to the bad evaluation. That is why i want to
>>work on that at the moment.
>That is unusual.
>You might analyze several failures against TSCP (with a very strong engine) and
>see if you can find a consistent point of failure (king safety/passed
>Then you might learn what needs fixing.
>Tactical strength does not imply playing strength.  In fact, I wrote a program
>that projected best fit parabolic curves through parameter sets and that would
>automatically tune for strongest tactical performance for Beowulf.  It made it
>stronger tactically, but overall it played much worse with the new parameters.

I do not know what is your definition for tactical strength but if it is ability
to solve faster test suites then it is wrong.

The problem is that test suite are not about tactical strength but about
sacrifices and a lot of tactics has no sacrifices in it.

It may be a quiet move that fail high not because of a sacrifice but because of
seeing deeper.


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.