Author: Will Singleton
Date: 00:52:19 07/12/99
Go up one level in this thread
On July 12, 1999 at 01:50:42, Peter McKenzie wrote: >On July 11, 1999 at 16:19:24, Pat King wrote: > >>I recently ran a series of tests on SweetReason 1.7, my chess program, turning >>on and off various features to determine their effect on search speed and >>program strength. In general, I got expected results. Ie, my program appears to >>gain 140 Elo points/ply, and the typically reported figure's about 200 (effect >>of increased knowledge in evaluation?). The surprise came in evaluating the >>check extension. Common sense would lead one (well, me) to expect the check >>extension to increase strength and decrease speed. I had exactly the opposite >>effect -- using check extension appears to drop 200 Elo points, but also reduce > >Yes, this is surprising! I have heard some people say bad things about check >extensions though. I haven't done any good experiments with them lately, >something else for the todo list... >Are you sure you didn't have a bug? > >>the nodes visited in a fixed depth search by 40%! I can write off the Elo effect >>to small sample size, but the other figure has enough data behind it to stand >>up. Could the improved evaluation of "checky" positions cause enough cutoffs to >>do this? > >what was your exact methodology? > >I assume you ran a bunch of test positions to a fixed depth? If so, which >positions? What depth? I would be interested in trying to reproduce this >result. > >peter Check extensions are well known to increase nodes-examined, as well as playing strength. This has been reported over and over again in the literature, and they are used by every program and programmer I've run into. So, you are probably doing it incorrectly. Will
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.