Author: David Rasmussen
Date: 04:32:09 01/27/01
Go up one level in this thread
On January 26, 2001 at 23:18:03, Vincent Diepeveen wrote: >On January 26, 2001 at 01:52:48, David Rasmussen wrote: >> >>It's still different from what I suggested. As Edward said: >> >>> >>>> >>>>applying david's suggestion to a null move implementation would >>>>mean reducing the search depth after a null move failed high >>>>instead of simply returning immediately with a fail high. >>>> >>>> - edward >> >>My idea is more general. > >Why do you waste all those nodes to reduce a single ply, >what you do is simply leading to incorrect search, as >depending upon alfa and beta you write to hashtable that a >depth of search depth = n is having an x score but somewhere in the >tree it actually searched because of a reduction n-1 ply. > >The reason why this probably this problem doesn't show much >as if it does in FHR is because if nullmove goes ok then >the whole rest of what you search in this position is no longer >relevant. > >so a huge speedup of your program is then to give a cutoff after >nullmove is > beta instead of searching on for the man with the >short name. > >Greetings, >Vincent I really don't understand what you are talking about here. I am not talking about FHR, or any other kinds of pruning based on alpha or beta. I am talking in general of searching those positions that seems inferior, less deep. If you don't like to base this on alpha or beta, then don't. But there are a million other cases where it might be useful. For example, as someone noted, we could decrease depth if the move we just made, immedeatly hung a piece, maybe based on SEE or some other criteria. This wouldn't depend on alpha or beta, and we would still see the brilliancy at the end of such a line, if it existed, eventually. If we can make such clever negative extensions, so that we gain an extra ply, then we might not loose tactical ability. It's the same sort of balance with positive extensions.
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.