Author: Dan Homan
Date: 05:12:05 02/11/99
Go up one level in this thread
On February 10, 1999 at 17:20:34, Dan Homan wrote: >On February 10, 1999 at 14:54:43, J. Wesley Cleveland wrote: > >>On February 09, 1999 at 22:59:52, Robert Hyatt wrote: >> >>> >>> >>>one of the big 'time savers' is what happens _below_ a sacrifice for nothing. >>>Qxp pxq. And now no matter what two moves white plays in a row, black's >>>position is still 'good' and everything below such moves gets trimmed away >>>instantly... >>> >>>IE with a full-width search, there are _many_ captures that are losers. Null >>>move reduces the depth below these moves and make dismissing them quick and >>>easy... >> >>This makes me wonder whether null move is even worthwhile if there is not a >>large material difference. I would think that with even material null move would >>be much less likely to produce a cutoff and much more likely to give a false >>cutoff. > >This is a very interesting observation. Maybe it would be right to >say that null move might be a "waste" of time if the material score >is not close to beta? Taken a step further... some kind of a "windowed" >null-move might be a good idea. > >I tried this in my program. I changed my null move code so that >a null move would only be tried *only* if the material score for the side >to move was greater than (beta-1.5*pawn_value). This reduced my >searched nodes from 5-10% in most of the positions I have tried so >far. (I have only tried a handful of WAC positions so far.) Ok, I tried the full WAC set. Overall the improvement is 0% to 5% depending on the position. Most positions seem to search a few % few nodes to reach the same depth. This is all with the 1.5 pawn window. In the end, I went from 277/300 in 5 seconds per problem to 278/300 in 5 seconds per problem. This is all on my 400 MHz celeron. - Dan > >If this pans out, It is a perfectly safe enhancement to >null-move. The 1.5 pawn window that I tried was arbitrary; >another value might do better. > > - Dan
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.