Computer Chess Club Archives




Subject: Re: Singular Extensions, Nullmove deepsearch

Author: Robert Hyatt

Date: 05:11:04 02/16/99

Go up one level in this thread

On February 16, 1999 at 04:45:43, Frank Schneider wrote:

>I experimented a little bit with nullmove deepsearch extensions and with
>singular extensions and now I have some questions:
>1. In all my tests the searchtree (#nodes to finish n iterations) increased
>   by a factor of 1.5 to 4. Although testresults were interesting I believe
>   this is too much. What are your experiences?

when I added singular extensions to Cray Blitz, the 'cost' was one ply.
which was about a factor of 5 in speed or nodes.

>2. Is it correct to assume that singular extensions work best if there are
>   only few other extension heuristics.

not at all, but you have to be careful as it is possible to over-extend and
drive the tree wild...  Getting out of check is not singular 99% of the time,
so discarding that would cause tactical trouble....  others would probably hurt
also.  IE in CB I didn't take anything out...  but you do avoid the singular
test in some cases (in check with only one legal move is obviously singular,
but I already extend for one-reply so the singular test doesn't happen.)

>3. How would you describe the difference between nullmove deepsearch extensions
>  (Donninger) and the nullmove threat detection by Deep Thought?

The ideas are similar, based on the result the null-move search returns.  I
found this not particularly good around version 8-9 in Crafty.  I tried
both singular and null-move threat extensions, and didn't like them back then.
But hardware has come a long way, so that giving up a ply now would not kill my
performance.  Singular extensions (at least a fairly easy to implement version)
is being tested right now by one crafty clone on ICC.

>4. Who uses
>   - singular extensions
>   - nullmove deepsearch extensions
>   - nullmove threat detection
>5. What is the best (easiest) way to implement singular extensions?

if you mean 'real singular extensions' there is only one way to do this,
as explained in the deep thought extension article.   It is non-trivial to
do right, otherwise you have to sort thru a lot of search inconsistencies
causes by a move extending at ply=n, but not at ply=n+1, etc..

I don't know of anyone using real singular extensions other than Deep Blue
and Cray Blitz (this being both PV-singular and FH-singular extensions).

>Thanks in advance,

This page took 0.13 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.