Author: Peter Fendrich
Date: 12:14:54 08/13/04
Go up one level in this thread
On August 12, 2004 at 11:48:35, Tord Romstad wrote: >On August 12, 2004 at 10:16:33, Peter Fendrich wrote: > >>On August 12, 2004 at 09:28:24, Tord Romstad wrote: -snip- >>I can't find out how to get almost_certainly_fail_high in my engine that doesn't >>evaluate interior nodes. There are a few endgame cases that I handle as >>known_fail_high even if they are (very) probably_fail_high. > >A simple addition which may or may not work for you is to do a preparatory >null move search with the depth reduced all the way to 0 before you do >the full null move search with depth reduced by R+1. If the first null >move fails low, you skip the second. > >Of course, this should only be done if the remaining depth is big. Yes, well worth trying! IIRC I have tried it but my memory isn't what it should be... I also had another thing in mind. Running Nullmove with a temporary alpha/beta. tempalpha = beta+margin and then in Nullmove use search(depth-X, -(tempalpha+1), -tempalpha) If that fail high we might take the risk of pruning or reducing depth. One thing is that the search will trash the hash table because of the new window so one shouldn't update it during this search. /Peter
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.