Author: Tord Romstad
Date: 06:25:27 03/26/04
Go up one level in this thread
On March 26, 2004 at 06:12:48, Roberto Nerici wrote: >Hello Tord, Hello Roberto! >>Yes, certainly. And I should perhaps have pointed out that I did not >>disable what I call "static null move pruning", but only the recursive >>null move searches. When the remaining depth is low or it seems very >>clear that the side to move is winning, and the static eval minus the >>biggest statically detected threat is above beta, I return a fail high >>score immediately. This makes the effect of the removal of recursive >>null move pruning less dramatic than it otherwise would have been. > >Is this "static null move pruning" actually doing a null move? It sounds more >like a futility-type scheme from your description. You are right, no null move is executed, but the idea is similar to the standard recursive null move search. I regard the null move search mainly as a technique for detecting threats. At nodes where the static eval is higher than beta, I would like to prune the entire subtree and return a fail high score immediately. My position is good, it is my turn to move, and I almost certainly have a move which is better than doing nothing, so why should I do any search at all at this node? In general, this is of course far too dangerous, because all sorts of nasty threats from the opponent would be overlooked. It is therefore wise to do some kind of threat detection before failing high. This is where the null move search enters the picture. It is a way to look for threats before pruning the subtree. If the null move search fails high, it means that the opponent doesn't even threaten to bring the score below beta, and it is almost certainly safe to return a fail high score. When the remaining depth is small, I've found it to be more effective to do static threat detection than to do a full null move search. A null move search is of course more reliable, but it is also much more expensive. >Btw, I've spotted a mistake on your web page for Gothmog. It says "on fast >computers it is probably good enough to offer the average chess player a good >match". Surely it should say "average master..."?! Possibly. The text is very old, and you'll find a lot of information there which is no longer up to date. The sentence you quote was probably rather accurate at the time it was written. >(You do seem very modest about your program!) Not really. I am just too lazy, and update my web page much less often than my engine. :-) Tord
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.