Author: Frank Schneider
Date: 00:06:10 07/12/99
Go up one level in this thread
On July 09, 1999 at 17:58:03, Peter McKenzie wrote: <big snip> >> >>In Paderborn, Chrilly openly admitted to employ multi-cut pruning. >>Maybe this is the "top secret" method ... >> >>The bibliographic refernces for multi-cut pruning follow below. >> >>Bj"ornsson, Y. and Marsland, T.A. (1999). >>Multi-cut pruning in alpha-beta search. >>1st International Conference on Computers and Games, >>Proceedings, H.J. van den Herik and H. Iida (eds.), >>pp. 15--24, LNCS 1558, Springer, ISBN 3-540-65766-5. >> >>Bj"ornsson, Y. and Marsland, T.A. (1998). >>Risk management in game-tree pruning. >>Technical Report TR 98-07, Department of Computing Science, >>University of Alberta. > >Perhaps someone could post a brief overview of what Multi-cut pruning is? >I've never heard of it before... > Ingvi sent me the first paper mentioned above. I'll try to explain the idea: - if you are at an expected cut-node (as opposed to an all-node) and not in check and there is enough material on the board - try to search 10 moves with a depth reduced by R=2 - if 3 of this 10 moves fail high cut without searching a move to full depth. Of course you can use different values than (3,10,2). The idea is that it is very likely that the node will really be a cut-node if 3 moves failed high. The idea can be described as a kind of 'inverse singular extensions' (use shallow searches to make sure a move is not singular). Very nice idea but so far I had trouble implementing it successfully. Frank
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.