Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: measuring search efficiency--ideas?

Author: Dan Homan

Date: 06:10:32 05/23/00

Go up one level in this thread


On May 23, 2000 at 03:59:50, Tom Kerrigan wrote:

>Hi guys,
>
>Is there a good way to measure search efficiency?
>
>In the past, I've gone through test suite logs and compared nodes/ply for each
>problem by hand. This is obviously undesirable. :)
>
>Is there some way I can get a good "magic number" to indicate how efficient my
>search is? Is branching factor good or bad? Is there something similar but
>better?
>
>Thanks,
>Tom

I don't know if this is what you are looking for, but a year or so ago on this
group, Bruce Moreland suggested comparing graphs of solution times and ply
depth reached as a measure of search improvements.  I liked this idea but
decided to just use mean ply depth and root mean square (RMS) solution time
instead.

The mean ply depth for a test suite is nice because it tells me about how deep
my program is going in a fixed period of time (I throw Mates out of the average
because they cut the search time short).  The RMS solution time is also nice
because I want my program to find the best move as quickly as possible.  I use
the RMS solution time because it weights up the problems that take the longest
to solve.

To optimize my search I try to get the best mean ply depth and RMS solution time
compromise on test suites.  If I turn up extensions, I usually can reduce the
RMS solution time significantly (particularly on a tactical suite), but the
mean ply depth goes down significantly as well.  If I increase pruning, my
mean ply depth goes up significantly, but my RMS solution time also goes up
(and some previously solved problems go unsolved).

I think the two measures are simple and off-setting which makes them useful
to me.

 - Dan



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.