Author: Dann Corbit
Date: 09:16:33 01/04/06
Go up one level in this thread
On January 04, 2006 at 06:14:05, Uri Blass wrote: >On January 04, 2006 at 04:33:37, Dann Corbit wrote: > >>On January 03, 2006 at 14:40:01, Tord Romstad wrote: >> >>>On January 03, 2006 at 13:22:57, Andreas Guettinger wrote: >>> >>>>On January 03, 2006 at 12:28:09, Robert Allgeuer wrote: >>>> >>>>> >>>>>>It is possible that Sergei introduced the name "history pruning", but the >>>>>>technique itself is very old; certainly much older than SmarThink. I no >>>>>>longer remember where or when I heard about it for the first time, but it was >>>>>>definitely not in this millennium. >>>>> >>>>>It would be really interesting where this technique came from, given that it is >>>>>now in wide-spread use. Maybe a forum member knows... >>>>> >>>> >>>> >>>>Two papers were it was introduced (1989), probably found on Dann corbits FTP. >>> >>>As Stuart points out, these papers are about history-based move ordering, >>>which is not the same thing at all. >>> >>>I don't think there are any papers to be found. History pruning/late move >>>reductions must rather be considered as a part of the "oral tradition", and >>>the origins seem to be lost in antiquity. Perhaps Bob or some other veterans >>>can tell us more. >> >>I would be interested in pruning data for difference based pruning. >> >>For instance, if some move is the root node, and has a centipawn evaluation of >>100 centipawns, then when a move fails low or has a hash table value of the >>appropriate depth at -100 centipawns, can we reduce the depth of search? How >>about 200 centipawns difference? 1000 centipawns? It is true, we will just have >>a bound, but if it is an upper bound and it is 200 centipawns below the pv node, >>can we reduce the depth of search by 1/2 ply? How about by 1 ply? >> >>The idea is sort of like null move, except that we scale reductions according to >>how bad the move looks. We could have an additional parameter for maximum >>reduction (e.g if we are supposed to look 12 plies deep, we will look at least 7 >>plies deep even if we lose a queen and a rook). I think that this would play >>more like most people. We tend to look harder at moves that look promising and >>not so hard at moves that look bad. > >Same for computers because of null move pruning. > >If you are queen and rook down you usually threats nothing. The thing that seems strange to me about null move pruning is the all or nothing effect. Losing a pawn is worse than doing nothing. So is losing a queen. But losing a queen is a lot worse. Yet both moves would get the same reduction.
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.