Author: Robert Hyatt
Date: 16:49:58 02/07/98
Go up one level in this thread
On February 07, 1998 at 18:48:18, Thorsten Czub wrote: >>I'm not so sure. Ed has said "no null move" for Rebel. I believe Marty >>said the same about Mchess. Looking at Hiarcs output, it seems that it >>doesn't either, based on the depth of search reported... The only one I >>am certain about if Fritz, of course, that is a known null-mover. Bruce >>and I are also big users of course. I've used it since Beal's first >>paper >>on the subject in 1980-81 or so... > >If null-move is so beuatiful and works so effective and without >overseeing important stuff, and gives more depth , why do you think is >rebel and mchess and hiarcs not using it ? >How can rebel9 and hiarcs6 lead the ssdf-list without null-move? > null move is a form of forward pruning, but not the only form. The advantage for null-move is that it is a "domain independent" form or pruning... it has nothing to do with chess and works just as well in other games. One simple comparison is current Crafty vs Blitz IV, circa 1977 or so. The code that selects moves to search (or, which moves to toss out, depending on your perspective) was over 30,000 lines of code. The search for Crafty, in contrast, is a few hundred lines at best. So the attractiveness, for me, is that it lets me write a simple, bug-free search, and get on to other things. It exhibits a failure here and there. So does Rebel and the others. My thinking is that if I am going to have to accept errors here and there, I'd just as soon do it with the simplest code I can. The problem with null-move is that it gives good results with forward pruning, but the "pruning" isn't controllable. IE If I want more depth, I can't start tossing out a few more moves easily. In traditional forward pruning, it is possible to produce the branching factor you want, so long as you are willing to accept the increased error rate more pruning produces. If you can control the errors, which is not easy, you could even beat null move pruning for efficiency. But there's nothing inherently bad about null-move searching. I did forget to mention the strongest program I know of that doesn't use null-move... deep blue... >(IF they don't use it - we have no evidence, or ?). > >Could we discuss null move, or is null move so SURE that we don't have >to discuss it ? And if it is so sure, why the hell don't they use it >?!!? I'm assuming they think what they do is as good or better. I'm not sure whether it is or isn't. But there's no huge difference for sure, as shown by Fritz in the Nunn position games... > >I would like to learn from you about null-move. How far can we trust it. >And despite trust, which negative effects can cause ed/marty/mark not to >use it (IF they don't use it - I am only quoting you). the biggest problem with null-move is the zugzwang problem... if you do reach a position where not moving is the best move of all, null-move will produce bogus results since not-moving is not an option in chess. I used to suffer when the opponent would get a bishop/pawn at f6, and queen at h6, and the null-move would reduce the depth enough that I would miss the Qg7 mate move since it wasn't a capture. As depth improved, and extensions were added, this problem has pretty well gone away. There are other tricks to let null-move be useful, even in single-piece endings where it can be dangerous.
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.