Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: testing of evaluation function

Author: Walter Faxon

Date: 23:11:06 04/17/03

Go up one level in this thread


As mentioned by the other responders, a good first-cut test for a static
evaluator is to compare evaluations over a test set of positions or games with
the evaluations made by good players or other programs (collectively,
"experts").  However, to make major improvements to the evaluator that is not
enough.

Due to the complex interaction of evaluation with search, your program will
often never even consider a position that your expert will insist is along the
main line.  Likewise your expert will often ignore what your program believes to
be the main line.

So, a suggestion:  have both your program and expert static-evaluate the deepest
quiet position along both main lines.  Although this is no panacea, differences
there should quickly show much of what your program is missing.

-- Walter



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.