Author: Daniel Karlsson
Date: 17:21:34 06/23/99
Go up one level in this thread
On June 23, 1999 at 18:06:10, Robert Hyatt wrote: >On June 23, 1999 at 15:46:25, Daniel Karlsson wrote: > >>I've been trying a new (at least it's new to me) move ordering heuristic and I >>wanted to see what the rest of you think about it. >> >>When I do a nullmove and it fails to produce a cutoff (on [beta - 1, beta]), I >>take the refuting move and use it sort of like a killer move for the next ply. >>This "nullmove killer" is searched before the other killers. This seems to >>reduce the tree by about 3% or so (7 ply search), but my program and hardware >>are too slow to do any exhaustive testing (or rather, I'm too impatient). >> >>Has this or something similar been tried by others? Was it any good? Is it plain >>stupid? Any comments appreciated. > > >It would seem to me that this happens 'naturally'. IF one move causes the >null-move search to fail low, then that move must have failed high and _must_ >be in the killer table from doing so? Well, ehm, I reached the same conclusion an hour or so after posting. The only difference is that I stored any move, even captures, in the "nullmove killer". It seems that on occation this helps, I suppose because the capture was sorted last as "bad" otherwise. Allowing "bad" captures in the killer list produced pretty much the same result. However in some positions I still get better results with "nullmove killers". Haven't figured out why and when yet. Oops!
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.