Author: Stuart Cracraft
Date: 09:21:01 09/23/04
Go up one level in this thread
On September 22, 2004 at 19:38:53, Robert Hyatt wrote: >On September 22, 2004 at 18:53:51, Stuart Cracraft wrote: > >>Are you saying I should consider marking each move after its search, then >>back up the score value, node count, etc., then resort based on that? >> >>I could see that various parts of the subtree have generated new history >>heuristic scores for from/to move coordinate pairs and that a resort could >>affect the remaining move order. >> >>I don't currently resort at any level after the movegen at that level. >>I just search all the moves in the original post-movegen-sort-order. >> >>I am fine with considering doing continual resorts but worry about the >>overhead and the return. But with it being the all important move-ordering, >>what have you seen in doing these resorts in terms of improvement >> >>Stuart > >There is _no_ overhead. It is done only at the root, once per iteration. For a >12 ply search, a total of 12 times. That won't use measurable CPU time. The >point is that root move ordering is critical for efficiency.. FYI -- I tried this for ply == 0, resorting based on the new history heuristic values after each ply 0 move is searched. My result went up slightly on problems solved and my total nodes searched drop by about 2 %. Stuart
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.