Author: Robert Hyatt
Date: 07:29:00 12/27/99
Go up one level in this thread
On December 27, 1999 at 00:07:47, William Bryant wrote: >If your search fails low on the root PV move. >(IE. the full alpha beta window for the PV move returns a score <= alpha >and nothing is backed up into the PV array). > >Do you? > >1) Stop PVS and search the rest of the root moves for this iteration with a > full alpha beta window? > >2) Stop PVS until at least one move for this iteration falls into the alpha > beta window. If no moves fall into the window, research the whole > iteration with a lower window. (fail low). This means that you may end up > searching every move with a full alpha beta window which would take more > time before the fail low occurs. > >3) Stop now and lower the window and research with PVS. > >4) Do Nothing? > Continue the rest of the root moves within the -alpha-1, -alpha PVS window. > Only search moves outside the PVS window for the rest of this iteration > if they return a score > alpha on the PVS search. Hope that if the > every root move is going to fail low, that this will speed up the > search until the window is lower. > >5) Suggest something better. > >Fail low on the root move seem to bring my search to a crawl and I'm looking >for a way to be more efficient. > >Thanks in advance. > >William >wbryant@ix.netcom.com I do 3. When the _first_ root move fails low, I drop alpha and search again. I want the score for that move as quickly as possible, so that I have some idea of how bad things are, and how much time I want to spend trying to fix it. I have tried two usual approaches: (1) as I do now (2) continue searching root moves with original alpha/beta window, hoping that another move will fall into the window and avoid a research with a wider window. I didn't like this because it is pretty common to "think" you are winning a pawn or something at shallow depths, but at deeper depths you find you are not. That fail low can take some time if you try _all_ root moves and all of them fail low... Which is better is hard to say. I have been using (1) for at least 20 years, although I have tried 2 from time to time but always discarded it... The thing I don't like about this way is that you _must_ search every root move (if they all fail low) and you cannot safely stop until you lower the window and go again, as the move at the front of the list might be horrible...
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.