Author: James Swafford
Date: 11:18:21 12/19/01
I'm just now getting around to implementing a null move search in the dinky little thing I've been playing around with lately. One of the move ordering stats I keep track of is the ratio of fail highs on the first mv vs. the number of total fail highs. Previously this number has been at 87% for the first 100 problems of the Win at Chess suite. (My eval is so simple I don't bother testing more than that yet. :) ) With a pretty kludgy null search with R=2, I'm getting a few more problems right, and searching a little deeper on average. An observation I made, though, is that the ratio mentioned above fell to 78% - a delta of 9%. At first glance that makes sense to me, though I'll have to think about it a little more later. It leads me to wonder, though, if this is consistent with more sophisticated engines. For those out there that keep track of that ratio, have you measured it without a null search? I'd be interested in hearing the numbers with and without the null search, just to be sure I didn't break something. Most folks know that a null move can break things in zug positions, so I'm left to wonder when it's a good idea to avoid it. In a previous program, I used a simple measure of material. I think I said if the material for the side to move was less than a rook, avoid the null search. Something like that. Anyway, has anybody come up with a better idea? I tried Vincent's "double null move" a couple years ago in Tristram, and I found it to be pretty good. Anybody else using it these days? (besides Vincent. :) ) -- James
This page took 0.02 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.