Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Test match with the Botvinnik-Markoff extension

Author: Anthony Cozzie

Date: 10:34:15 10/02/03

Go up one level in this thread


On October 02, 2003 at 13:07:52, Uri Blass wrote:

>On October 02, 2003 at 10:56:21, Tord Romstad wrote:
>
>>On October 02, 2003 at 08:09:41, Anthony Cozzie wrote:
>>
>>>On October 02, 2003 at 05:21:45, Tord Romstad wrote:
>>>
>>>>I have just finished a 100-game test match between a version of Gothmog with
>>>>the Botvinnik-Markoff extension and an identical version without the extension.
>>>>The version with the new extension added won by 56.5-43.5.
>>>>
>>>>Not enough data to make any definite conclusions, of course, but it certainly
>>>>looks interesting.
>>>>
>>>>Tord
>>>
>>>Could you post some positions where BM-Gothmog was able to outthink ordinary
>>>Gothmog?
>>
>>Unfortunately I haven't studied any of the games -- I just left the match
>>running overnight, and checked the result when I came back to my office
>>this morning.
>>
>>The only position I can offer you now is WAC141, which is solved in
>>8 plies (about 6 seconds) without the extension, and in 7 plies (about
>>1.5 seconds) with the extension added.
>>
>>>Just offhand, it seems to me like this extension would extend a lot of
>>>useless things, but your result seems to indicate otherwise.
>>
>>It is not terribly expensive, at least not with my implementation (I only
>>extend when the threat is rather serious).  In non-tactical positions, the
>>slowdown is hardly noticable.  In tactical positions, I need about 10-20%
>>more nodes to search to the same search depth, but apparently this is
>>outweighed by the fact that a few tactical shots are seen one ply earlier.
>
>You implemented it very fast.
>I think about implementing it but it is not a trivial task.
>
>I understand that you use the move that comes after the null move to determine
>the target.
>
>The problem is that after undoing the null move I do not know the move
>that came after it in case of no moves after it and it is possible that the
>qsearch that came after it generated no move.
>
>I can determine a special varaible nodesnull and have nodesnull=nodes after
>null move and check if nodes>nodesnull after I undo null move and check for a
>threat only in case that nodes>nodesnull.
>
>I wonder if you did something similiar.
>
>Uri

If I understand it correctly, implementing the BM extension in Zappa would go
like so (pseudoish):

   do_null_search();
   rb->null_threat = rb->best_move;
   if(ply >= 2 && m_f_loc((rb-2)->null_threat) == m_f_loc(rb->null_threat))
       trigger_extension();

anthony



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.