Author: Robert Hyatt
Date: 09:10:27 07/01/04
Go up one level in this thread
On July 01, 2004 at 11:38:44, Steve Maughan wrote: >Monarch, like many engines, suffers from search instability i.e. while search a >move it fails high, only to then fail low when the move is searched with a wider >window. I've notices quite a few tactical positions where Monarch "find" the >key move only to lose it again on the research. Clearly this is not good and >could be due to bugs. However I also notice that many of the top engines have >extremely stable searches. So here's the question - what general or specific >things can I do to improve the stability of Monarch's search? > >Thanks, > >Steve If you are talking about PVS, where you search on a null-window and fail high, then re-search on a normal window and fail low, the fix is easy. If you fail low, _ignore_ the fail high, keep the old best move, and keep searching. If you mean you fail high on the null-window search, relax beta and fail high again, and now return to the root and increase beta significantly and fail low, that fail-high can be trusted and kept. In short, if the PVS null-window search fails high, don't trust it unless the verification search produces a good score or another fail-high. That will stop all the nonsense.
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.