Author: Robert Hyatt
Date: 02:10:37 08/03/98
Go up one level in this thread
On August 03, 1998 at 04:10:34, Guido Schimmels wrote: >Getting a fail-low at the root is annoying, as we don't know what move to play >for quite a while. >There is in article by Schaeffer et al. ("Advances in alpha-beta searching") >suggesting to restart the search in these cases (setting depth to 1 again), >exploiting the benefits of a hash-table and finding a new best-move quickly. >But nobody seems to use this, why ? (or am I wrong here ?). >An important implementation detail not mentioned in this paper is, how to do the >windowing up to the search depth we restarted from - if we do aspiration as >normal isn't there a danger to get caught in an eternal restarting-loop ? > >- Guido - I haven't seen this, but don't see how it would work. Because it took the search reaching depth D to fail low in the first case. So all that will happen on the re-search from depth=1 is that you will fail low, then reduce alpha to -infinity and search again. But now the hash entry that says "fail low" can't be used, so you fail high or else get a valid score that is back in the vicinity that it was when this mess started. If you have to reach depth=D to fail low, searching any more at depth D-1 will not work, because it won't understand the position (won't have enough depth or extensions) to find the reason for the fail low. So there could be a danger in looping, but in reality, there is a danger of just wasting time, rather than trying to find a better move to replace the one that failed 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.