Author: Joachim Rang
Date: 06:49:53 09/18/03
Go up one level in this thread
On September 18, 2003 at 09:25:08, Steven Edwards wrote: >On September 18, 2003 at 04:57:02, Peter Fendrich wrote: >>On September 18, 2003 at 04:18:35, Steven Edwards wrote: > >>>The programs Kaissa, Caps, and Paradise all had a causality facility. Such a >>>feature is used for forward pruning in a manner suggestive of the human move >>>selection process. The basic idea is: >>> >>>Given a subtree search of a position P with move M1 from P, the causality >>>facility produces a causal data structure that contains the various constraints >>>on the position and the move that allow the evaluation returned from the >>>subsearch to be valid. Forward pruning is achieved when a move M2, a sibling of >>>M1, can be applied to the causal data structure with the result that M2 doesn't >>>violate any of the constraints and so doesn't need to be searched. >>> >>>It's not an easy problem, and some solutions may require more resources to >>>calculate and probe the causal data than to do the unpruned search. >>> >>>Are there any current programs that have a causality facility? > >>During the search of M1 the program collects some values=constraints and builds >>up a causal data strucuture. That structure is used during the search of M2-Mx >>(siblings of M1). Right? > >Yes. > >>What kind of constraint values could that be? >>How does that form a data structure? >> >>Do you have some example values... > >D] rnb1kbnr/ppp1pppp/8/3q4/8/2N5/PPPP1PPP/R1BQKBNR b KQkq - 1 3 > >A program is playing Black with the above root position as position "P". The >first move on the list, M1, is a6. The search tries this and gets a poor score >and a causal data structure. The program applies the next move M2 on the list, >a5, to the data structure and sees that the constraints that produced the bad >score for M1 also produce a simisar score for M2, so M2 is not searched. > >Some of the constraints in this case: > >1. The wN/c3 remains there, unpinned, and able to capture bQ/d5 > >2. The bQ/d5 remains there and is undefended > >3. Black has no known counterattacks or other resources within the depth of the >subtree search. I think with a killer-slot for move-ordering all moves which left the black queen on d5 will be dismissed very soon. No need for a causility facility. I think Rebel does some interesting form of forward pruning based on evaluatuion. It seems to me, that this is comparable to your approach. I also made a post concerning forward pruning based on evaluation, which you can find here: http://www.chess-archive.com/ccc.php?art_id=306821 regards Joachim
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.