Author: Robert Hyatt
Date: 18:12:46 09/24/98
Go up one level in this thread
On September 24, 1998 at 16:11:54, John Coffey wrote: >It seems to me that there would come a point where attempts to order >the moves at every branch of the tree would slow down the tree search. >I could see doing this in the lower levels of the tree, but wouldn't >there be a point of diminished returns? I.e. if you were searching >N ply deep, maybe you would want to stop ordering at maybe N-3? > >This brings up another issue: Crafty (and for all I know, other programs) >will generate the entire move list at each branch before searching deeper. >It might do some ordering too. But it might not be necessary to look at >all these moves, so the time spent generating some of the moves could be wasted. > >My plan, right or wrong, for my Mars engine was to have different search >strategies at different levels of the tree. Probably most programs do this >already. > >John You are overlooking a serious issue... the tree search grows *exponentially* with depth. Which means that any effort expended to order the best move first at any ply will greatly pay off in reduced tree search space. No matter how deep in the tree it is, capture or regular search. With alpha/ beta, move ordering is not just an issue, it is *the* issue. You can *easily* make your tree 10X larger with sloppy move ordering, and you will be faster than everyone in terms of NPS, but you will be searching 2-3 plies less deeply than everyone else. It is the *depth* and not the *NPS* that is important...
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.