Author: Robert Hyatt
Date: 08:02:07 04/23/01
Go up one level in this thread
On April 23, 2001 at 09:45:45, Ulrich Tuerke wrote: >On April 23, 2001 at 09:34:20, Robert Hyatt wrote: > >>On April 23, 2001 at 05:52:00, JW de Kort wrote: >> >>>Dear friends, >>> >>>I have added a qsearch to my program and as a result the number of nodes in my >>>search has exploded. Sometimes the program searches ten times as much nodes in >>>the qsearch than in the regular search. Can anybody give me some advice on >>>methodes i can use to limit the number of nodes in the qsearch? >>> >>>Thanks in advance. >>> >>>Jan Willem >> >> >>A simple idea works well: >> >>When you enter a q-search node, if the current score is (say) a queen below >>alpha, then capturing a pawn or piece is not going to bring the score back up >>to alpha. Those captures are useless to examine.. > >I think that this is just what I had suggested in my reply, but you have >explained far better why this should work. > >You say "useless to examine" . After all, this algo involves some kind of >forward pruning and there is a certain risk in missing some tactics. > >However, I admit that this trick appears to be reasonably safe. I guess, >everybody is doing it. > >Nevertheless I had observed that in some rare cases, a test position will be >solved later (i.e. in a higher iteration) as a consequence of this trick. > >Do you agree ? > >Uli Yes. And this is why I was so surprised to find that Hsu was doing this in Deep Blue 2. He was a fanatic about "accuracy" and the idea of not following some captures really went against his nature. But he mentioned that "futility" pruning in the q-search was one thing they did in the second chip (for the 1997 match) that was a big winner... It will miss things here and there. But it will also find things faster in more cases.
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.