Author: Robert Hyatt
Date: 11:38:03 11/01/00
Go up one level in this thread
On November 01, 2000 at 13:35:39, Ernst A. Heinz wrote: >>I am going to run this experiment myself one of these days, just to get a hard >>number for myself. 20-30% seems too low, IMHO, unless the test positions are >>tactical and easy, like WAC. If you find the right move quickly, then of course >>you search the optimal tree (or nearly so). The right test is a series of >>consecutive positions from a game, without a clearly "right" move. > >Tony's figure summarizes the results on all BK positions. > >Please also remember that these numbers always compare >"PVS + good move ordering" vs. "critical alpha-beta tree" >for fixed-depth searches of uniform path length. So, we >essentially measure the effectiveness of PVS combined with >good move-ordering schemes as opposed to the bare quality >of the move ordering in isolation. > >=Ernst= That is one too many degrees of freedom. I am interested in the case "given the current Crafty search/evaluation, how much smaller would the tree be if move ordering were optimal?" Or re-phrased to the original post in this thread, "how close is the typical tree searched by Crafty to the minimal tree that would be searched if move ordering were perfect?" Adding PVS to one but not the other doesn't make much sense, IMHO. It is a well-known approach that tries to take the search closer to the minimal tree size. I'll make it a point to run this test and report the actual tree size difference between normal and perfect move ordering, with everything else _exactly_ the same.
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.