Computer Chess Club Archives




Subject: Re: testing of evaluation function

Author: Robert Hyatt

Date: 06:26:33 04/18/03

Go up one level in this thread

On April 18, 2003 at 08:07:41, Frank Phillips wrote:

>On April 17, 2003 at 21:58:47, Robert Hyatt wrote:
>>On April 17, 2003 at 11:28:44, Steven Chu wrote:
>>>I have an evaluation function in my chess program and need to be able to test
>>>it.  I have been told to get some things to test it with and note down the
>>>desired output and the actual output.  The thing is that I don't know how to go
>>>about testing the evaluation function.  What sort of things do I use to test it
>>>and how do I do it?
>>>If anyone can help I will be very grateful
>>This is not "easy" but I have a piece of code that will suck in a FEN position
>>and call Evaluate() and print the score.  I put a bunch of positions in a file,
>>and run the "evtest" command.  I then look at the positions, one by one, noting
>>Crafty's static evaluation.  If I disagree, I flag it to look at.  If not, I
>>keep going.
>>It is good to catch oddball cases that trigger unexpectedly large scores for
>>the wrong reasons.
>I also reverse the FEN of each position to check that the white and black score
>is the same, which has identified a few bugs.

I added a flip and flop command.  I reverse it interchanging a1 and a8, etc.  I
flip the
board changing a1 and h1, and I finally do both, interchanging a1 with h8.  And
then I
make sure that all scores are identical.

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.