Author: Inmann Werner
Date: 06:38:54 12/14/98
Go up one level in this thread
On December 14, 1998 at 05:34:41, Andrew Williams wrote: >Hello all, > >I'm trying to work out how to implement capture/recapture extensions in >my program. I have had a look at crafty and gnuchess and I've tried >several approaches, viz: > >1 Extend when a capture followed by recapture restores the material > balance we had at the root. > >2 Extend when the capture/recapture restores the material balance from > before the sequence. > >3 Extend when the capture and recapture are on the same square (and same > material value). > >In all these cases, I seem to get a much larger tree and often lose a ply >of search because of this. So, three questions: > >(a) Am I on the right lines with 1, 2 and 3 above? >(b) Is it normal to lose depth when doing this? >(c) Generally speaking, it better to complete d7 with this extension than > d8 without it? > >I realize that these questions probably aren't yes/no style questions, so >thanks to anyone who takes the trouble to reply. > >Regards > >Andrew Williams >(Author of PostModernist on ICC) Hello Andrew! Nice game yesterday on ICC, although Inmichess lost :( Your questions: (only from my sight of the world) I extend always, when there are captures/recaptures to the same field. The only thing behind it is, that I want to assure, that the most time interesting capture, recapture sequences are fully calculated, and not only in quiet search. So I do not see, why you look at material balance or same pieces. If the material balance gets out of hand, normaly alpha-beta generates a cutoff. To not enhance too much, I set a "enhancement limit". All enhancements together are not allowed to break it. (I personally set it to 5) It is clear, that the search tree expands here. On the other hand, I try to expand, where I think there are interesting lines, and so on this lines I get a much deeper search. And that is, I think, the intention on extensions. The most interesting extensions are sure the check extensions. Here you also extend, cause you think, it gets interesting. The most winning in strength, I got when I extended on null move fails. Here I find fast mate threats (2 plys earlier than in normal search) and extend there. Nice thing. Making extensions is always difficult. It increases the search tree. The question is, if you really hit the interesting positions, or only extend on senseless branches most time. And this really is difficult to decide. Werner
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.