Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Looking for a last moment operator for Olithink 4.1.3 for CCT-6

Author: Tord Romstad

Date: 15:01:50 01/28/04

Go up one level in this thread


On January 27, 2004 at 15:42:27, Dann Corbit wrote:

>On January 27, 2004 at 14:40:37, Tord Romstad wrote:
>
>>Hi Dann,
>>
>>Are you sure it is a good idea to re-search if the first IID search
>>fails low?  I have never tried this myself, but at first sight it
>>doesn't look like a good idea to me.  When the reduced-depth search
>>fails low, it is likely that the full-depth search will also fail low.
>>At fail low nodes, move ordering doesn't matter much, because you have
>>to search all moves anyway.
>
>Because it is used for move ordering, and because the depth is 2 plies below the
>current search depth, it is a sure win.  The cost is 1/400th of a full search,
>and the win is a good guess for the pv node.
>
>At least when I tested it in Beowulf it worked well.  But I did not try it both
>ways in Olithink.

Very interesting.  I always assumed that doing another search when the
first reduced-depth search fails low was a waste of time, and never
even experimented with anything else.  It is probably worth a try for
me, too (though it wouldn't surprise me if this is one of the many things
which are very different in MTD(f) compared to PVS).

In a way, one could say that you and I use the opposite extremes of the
spectre.  I *never* do another search if the first search fails low, while
you *always* do another search.  It is possible that the optimal approach
is somewhere in the middle.  If the reduced-depth search fails by only a
small amount, do a re-search, otherwise, drop the re-search.  There's
lots of scope for experimentation.

Tord




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.