Author: Robert Hyatt
Date: 16:48:03 09/15/04
Go up one level in this thread
On September 15, 2004 at 18:42:50, Gerd Isenberg wrote: >On September 15, 2004 at 18:30:56, Gerd Isenberg wrote: > >><snip> >>>>>I guess with a counter of played pv-moves (0..N) in the top of the current path >>>>>and ply index one may determine the minimal tree node type as well... >>>>>So we have distance D of the current node to the last pv node in the current >>>>>path. >>>>> >>>>>If D == zero we have a pv-node. >>>>>If D is odd we have a cut node. >>>>>If D is even and greater zero we have an all node. >>> >>>That is crude. The problem is that suppose you are wrong at ply=2. That causes >>>you to mis-categorize every node below that because when you are wrong at ply=2, >>>for non-PV ply=1 moves, it suggests that either the current ply-1 move is better >>>than the best so far and you are changing your mind, or that the current ply=2 >>>move is not the best and another move will cause a cutoff. But below that ply=2 >>>move that is normally a CUT node, ply=3 would normally be an ALL node, but that >>>changes since the ply=2 node is wrong. > >i have your point, but anyway that are probably nodes of some interest, if the >formal node categorizer contradicts eval (or even reduced search) bound compare >in some way. That was what I spent a lot of time on in the DTS "where to split" code in CB. I first did a simple analysis to type each node, going root to tip, then I looked for inconsistencies from tip to root to see if I could pick up on where I blew move ordering which means splitting the tree in that area is dangerous...
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.