Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Looking for test positions

Author: Brian Richardson

Date: 08:01:53 03/14/00

Go up one level in this thread


I was in the same situation a few months ago and found the BT, BS and LCT tests
a little intimidating since my program did rather poorly at first (and still
doesn't do all that well--although I was a little surprised at how accurate they
have turned out to be given Tinker's ICS ratings).

I would suggest starting with the WAC300 suite.  I run numerous passes and have
developed some idea of what to expect at different time limits.  You may want to
start at 1 sec per position, then try 10 secs.

Once you get the feel of YOUR program you could certainly use the other suites
too, but they run with much longer time controls 900 or 600 secs and usually
take overnight.  I have also picked 5/6 positions that I have built into the
program and can test very quickly to make sure I don't make things much worse.

The null move R depth reduction, extenstions, killers, hashing, and various
other search factors will also have a big impact, in addition to just your eval
function changes.  I would tune the "time to a fixed depth" for some test
positions in your search routine first and then "tinker" with the eval function
after you feel you have a good search option mix.  Some tactical test suites are
more easily solved with R=2 and others with more aggressive extensions (check,
pawn, etc).

FYI, for WAC300 Tinker solves about 253 at 1 sec and about 280 at 10sec (on
pretty fast hardware PIII 733MHz and 128MB hash).  Also, some problems are or
are not solved depending on the size of the hash tables (from the FINE suite).

Good Luck!

PS When you are comfortable, put your program up on ICS and you will learn a lot
there too.



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.