Author: Simon Finn
Date: 13:05:36 09/20/01
Go up one level in this thread
On September 20, 2001 at 13:45:34, Robert Hyatt wrote: >On September 19, 2001 at 05:33:40, Uri Blass wrote: > >> >>I believe that Crafty has access to the five piece tablebases. >>The fact that the position in the end of it's pv is a tablebase position does >>not contradict it. >> >>It is known that programs do not use tablebases everywhere in the tree(otherwise >>they are going to be slower) and crafty does not use tablebases in the qsearch. >> >>Uri > > >I always worry when I see a capture delayed until the q-search. This usually >means one of two things: > >1. The resulting position after the capture is a draw, while the engine scores >suggest that the position is better than a draw. The engine defers making that >capture until the q-search where the egtb score won't be found and the draw is >"avoided" in a bizarre way. The position after the capture in the (rather strange) PV looks like a win. My guess is that there's a subtle variation going on here - the engine score at the end of the PV may be better for Black than the engine score in another (drawing!) pawn capture line, so Black chooses this (losing) line instead. The 5-man database don't help as much as we would like here because White can always play enough "nothing" moves to push the capture into the q-search and force Black to choose the wrong line based on inaccurate engine assessments. Simon > >2. The capture simply can't be seen in the normal search, which means that it >may or may not be a won position. Searching another ply would reveal the >answer when that moves the capture out of the q-search to the end of the >normal search. > >I have seen some _very_ creative horizon effects caused by trying to push the >capture beyond the egtb probe limit.
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.