Author: Robert Hyatt
Date: 16:09:59 06/17/02
Go up one level in this thread
On June 17, 2002 at 12:35:04, Gian-Carlo Pascutto wrote: >On June 17, 2002 at 11:23:27, Robert Hyatt wrote: > >>Crafty doesn't clear the hash table between moves. With a couple of >>exceptions... If something significant changes so that some scoring terms >>are updated, then the _scores_ are marked invalid in the hash table. But the >>best move is kept. >> >>fail high and fail low happens regularly. Just look thru a log file and >>notice all the "++" lines. Look for a ++ line followed by a -> line which >>says the search failed high, then failed low, and the iteration completed. > >Ok thanks, I was looking for '--' to see if you had fail lows. > >How do you handle the case if it happens? Full-window research? >Ignore fail-high or fail-low? > >-- >GCP I don't handle it very well. I just accept it and keep going. And at the next iteration I find that the score is lower than I had hoped, but the move _must_ be the best move as the hash table said so, even if I can't prove how much better it really is... I ignore null-window fail highs if they immediately fail low on the re-search, but if it isn't a null-window search (and what we are talking about isn't) then I just accept the result and keep on searching...
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.