Author: Robert Hyatt
Date: 08:59:40 01/29/04
Go up one level in this thread
On January 29, 2004 at 11:00:52, Thomas Mayer wrote: >Hi Bob, > >>> Really ? I think I disagree. When this happens at the root you don't accept >>> the failhigh score, so why would you inside the tree ? > >> If I fail high on the aspiration window at the root, I _know_ that is a valid >> fail-high. And I _always_ accept that as the best move no matter _what_ >> happens on the re-search. > >err... I might misunderstood this - but do you have no handling of fail-low in a >fail-high ? I have seen that at least Fritz accepts the fail high always and >only research when another move in the same iteration seems also to be a fail >high - else he switch after having searched all other moves immediately to the >next depth... Is it the same in Crafty ? > >Greets, Thomas Let me explain it, hopefully better. At the root, there are two kinds of fail highs that can happen in a PVS program. 1. I can fail high on the original aspiration search window. IE depth=N says the score is .5, when I start depth N+1, I set the window to (roughly as this is adjusted by the "odd/even" issue a little) to .3 and .7 (or whatever aspiration window you might want to try instead). This (N+1) iteration search might fail high as the true score is greather than .7. When that fail-high happens, I _always_ accept it. No exceptions. 2. I get the score for the first move normally, then start doing a PVS search on the remaining ply=1 moves. If I get a fail high here, I first re-search with the original aspiration window rather than the alpha,alpha+1 null window. If that fails high or returns a true score, I use it. If it fails low on this re-search, I throw it away and keep the original best move and score as best. This is most often caused by null-move failures deeper in the tree. So on a fail high on the first move, I accept it. An aspiration window fail high on any other move is instantly accepted. A fail-high on a null-window (PVS) search is not accepted until the re-search avoids failing low.
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.