Author: Robert Hyatt
Date: 09:14:57 10/08/02
Go up one level in this thread
On October 08, 2002 at 10:58:12, Rolf Tueschen wrote: >On October 08, 2002 at 10:48:52, Robert Hyatt wrote: > >>On October 08, 2002 at 08:04:58, Vincent Diepeveen wrote: >> >>>On October 08, 2002 at 00:52:38, Eugene Nalimov wrote: >>> >>>thanks for your explanation Eugene, >>>it's very consequent to what is mentionned in the paper. >>> >>>Note that I'm amazed he's still defending not using nullmove. In 1997 >>>i remember the many postings from bob saying how dubious nullmove >>>was. Nowadays with more powerful processors reality has proven >>>otherwise. >> >> >>No it hasn't, really. There are two ways to accomplish this "thing". You can >>reduce the >>depth on uninteresting moves (null-move) or you can increase the depth on >>interesting >>moves (selective extensions). The two approaches are theoretically >>_identical_... >> >> > >Could you explain that in common language? Are you sure? Identical? > >Rolf Tueschen Think about it for a minute. Let's play a position where each side has just two moves at every point in the tree. Your null-move algorithm will discover that one of the two moves can be (safely) evaluated to a depth of two plies less than the other. That is what null-move does, reduce the depth by 2 plies and if the result is still good, we take it and run. My selective extension algorithm does the opposite. It extends any move by two plies that is "interesting". Now define "interesting" as "any move that null-move search will not fail high on." In this case, any move you search two plies less on, I won't extend. Any move you have to search to full depth, I extend 2 plies. We search _exactly_ the same tree. Which is the point... You claim a depth of N plies for your search. I claim a depth of N-x plies for my search. You are "out-searching" me on depth. Our trees are _identical_. Which is better? If you like to report huge depths, null-move looks better. If you only care about actual results, they are equal. If you want to report "low depths" and be thought of as an "intelligent/slow" program, then you use selective extensions. :)
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.