Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: measuring search efficiency--ideas?

Author: Bruce Moreland

Date: 10:47:15 05/23/00

Go up one level in this thread


On May 23, 2000 at 13:28:45, Tom Kerrigan wrote:

>This is usually what I do by hand but it makes me a little nervous. It's
>possible that a program searches to 7 ply quickly but 8 ply slowly. So if you
>make the program a little slower, you don't finish 8 ply, so you compare the 7
>ply scores, and all of a sudden the program appears faster (even though it's
>actually slower).

This is a defect in my strategy, certainly.  That's why I don't do it for just a
few positions.  I do it for all of LCTI, for instance, which is 28 positions.  I
can look at individual positions and figure out what is going on, or I can look
at the total for the whole suite.  I think it's close enough.  It doesn't work
very well on the SMP version though, because I can run the same version twice
and get very different results.

>What I've started to do is search all the BK positions to n ply and count the
>total nodes. I think this is a good idea.

This would be a good measure.  I do time for my tests, because if I add to my
eval function I might get more cutoffs but the eval function might take more
time to execute, so I get fewer nodes and more time.

bruce



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.