Computer Chess Club Archives


Search

Terms

Messages

Subject: ECM and extensions

Author: Scott Gasch

Date: 14:30:55 01/08/01


Hi all,

I've been running my engine vs. the ECM 879 position test suite recently and
trying to improve its score.  Currently it gets about 500/879 correct at 20 sec
per move.

I was reading in Ernst's papers about extended futility pruning and noticed that
he includes DarkThought's results on the ECM to 8, 10 and 12 ply.  His engine
DarkThought scores more positions correct on the ECM even when searching to a
fixed depth of 8 ply than my engine does searching 20 sec / move!  My engine is
getting 9-10 ply deep in most of these positions at a 20 sec time control and
would have expected better results.

This leads me to believe DarkThought must have some smart extensions in search.
I currently do check, nullmove threat and pawn push extension.  I am thinking
about a recapture but do not know how to perserve a small tree size with this
extension turned on...  I see Bob's method for limiting recapture extensions and
the method used by the author of AMY... does anyone have other ideas on how to
keep tree size in check with recapture extensions?

Second, does anyone have any other intelligent extensions to discuss?  Is my
program's poor tactical performance on the ECM test suite a function of the lack
of recapture extension or it there some other extension missing?

As always, appreciate the help,
Scott



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.