Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: question to Tord.

Author: Tord Romstad

Date: 05:02:45 04/11/04

Go up one level in this thread


On April 11, 2004 at 06:33:10, Daniel Shawul wrote:

>Hi tord
>
>what does gothmog selectivity 6 means?
>My guess is you do null_move for first 6 plies and
>static eval pruning for the rest. I mean is it like Ed's?

No, it's not like Ed's at all.  It hasn't even got anything to do with plies.
The 'selectivity'
parameter controls one of the criterions which are used to determine whether a
move
should be searched to reduced depth.  The first three moves at all nodes in the
main
search are (almost) always searched to full depth.  All remaining moves are
candidates
for reductions.  For each move, I first check whether it is a capture, check, or
some other
kind of interesting move.  If the move does not appear to be interesting at all,
I do one
final test before I decide to reduce it:

  if(History[Compress[m->from]][Compress[m->to]] <
     (Nodes >> (20-EngineOptions.selectivity)))
    return 1;

As you can see, I compare the history counter of the move (i.e. the number of
times
the move has failed high in the past) to the number of nodes searched.  If the
move has
almost never worked before, it is also not likely to work this time.
EngineOptions.selectivity
is a variable which contains the selectivity setting in the init file.

When a reduced move fails low, I search it again with full depth.

>I saw in gothmog's ini an option to turn off null move.
>Well does that mean selectivity 0 , which does only static eval
>pruning for the whole of the iteration depth.

Turning null move off in the init file completely disables dynamic null move
pruning.

>I have added static selectivity pruning in DanChess.
>This is the first time i disabled the null move code in search of other
>methods . Guess what it is not that bad! It takes around 330000 nodes to reach
>Depth 9 at the initial position for both methods. True,the static method
>sometimes make clear mistakes at positions where positional score dominates.
>The static method is only at embroic stage and that is expected any way.
>since you are more experienced than me in these method,i am hoping you will
>help me with ideas.

I am not sure I have much more experience with you about this.  I do something I
call "static null move pruning" when the remaining depth is less than 3 plies,
but
not further away from the leaves.  Most of the selectivity in my search is in
the form
of extensions and reductions rather than in complete pruning of subtrees.

Of course, I will still help you as much as I can, and I wish you good luck.
:-)

Tord



This page took 0.01 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.