Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Threat Extension and Tree Explosion

Author: Peter McKenzie

Date: 15:37:59 05/23/02

Go up one level in this thread


On May 23, 2002 at 12:53:40, Miguel A. Ballicora wrote:

>On May 23, 2002 at 03:40:29, Steve Maughan wrote:
>
>>I have recently been tinkering with threat extensions (TE).  I'd definitely like
>>to include some form of TE but I have encountered some problems of tree
>>explosion.
>>
>>What is happening is that the null move routine is detecting a mating threat and
>>I'm extending by one ply at ply[x].  I then try a move (at ply[x]) that doesn't
>>aviod the threat. In reply the opponent plays a sub-optimal move at ply[x+1]
>>that does not lead to mate even though a forced mating move does exist.  At
>>plt[x+2] I then detect a threat and extend again...
>>
>>This sequence leads to a tree explosion.  Is there any common wisdom as to how
>>to avoid this?  Some ideas that I've had are:
>>
>>1) Only extend by a fraction - inelegant solution IMO

I wonder why this is inelegant?  I would say the opposite.

>>2) Store the rely to the null move that gave the checkmate and make this the
>>Killer move for the next ply - didn't seem to work well - still some tree
>>explosion.
>>
>>Has anyone any ideas?
>
>I do not have explosion when I do something similar to what you say, but I do
>not know why. I think that I do what andrew suggests, I limit the number of
>extensions in a line (I'll have to check). Also, I guess that the killers should
>take care of most of the problems.
>
>Are you sure that the killer is included in the right slot? i.e. if white
>killers are in "even" indexes (killer[2], killer[4] etc.) black killers should
>be in odds indexes (killer[1], killer[3] etc.) even if you do a nullmove.
>In other words, when you do "nullmove" make sure that you increase the "ply"
>counter too. Othewise, you end up storing a white killer when it is black turn.

Yes, this is a good point.

Perhaps killers could be given an extra high ranking if we detect this
situation?  Eg. following a nullmove which indicated a mate threat.

>
>Regards,
>Miguel
>
>>
>>Thanks,
>>
>>Steve



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.