Author: Robert Hyatt
Date: 12:58:15 09/05/02
Go up one level in this thread
On September 05, 2002 at 15:33:39, Uri Blass wrote: >On September 05, 2002 at 14:06:08, Eugene Nalimov wrote: > >>Actually, often you don't want to search the objectively best move first. You >>want to search the move that will cause a beta cutoff and will result in a >>smallest subtree being searched. >> >>For example, if you are currently ahead in a material (compared to beta) than >>you probably don't want to start a deep sequence of mutual checks. All you need >>is some quiet move that will preserve your advantage. > >It is not clear >I may prefer also a check if I suspect that the check may be a checkmate. > >I agree that you do not want to search the best move first >and I also believe that the 92% is not correct because in 92% of the cases that >there is a good move the first move is good enough to give a score above beta >but it does not mean that the order cannot be improved. > >The only good test for order of moves is simply time and I do not see a simple >way to calculate the theoretical maximal gain that it is possible to get by >better order of moves. > >Uri In parallel search that 92% is _critical_. It means that in 8% of the fail-high positions, the first move does not produce the cutoff and a parallel search can be started there in error. That error dwarfs the tree-size issue completely. The higher that 92% goes, the more efficient the parallel search becomes.
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.