Author: Robert Hyatt
Date: 07:20:38 01/06/01
Go up one level in this thread
On January 06, 2001 at 06:45:52, David Rasmussen wrote: >On January 05, 2001 at 23:48:58, Robert Hyatt wrote: > >>>> >>>>One problem if you do null-move, is to watch for cases where a root move fails >>>>high, then fails low on the re-search. In that case I do _not_ make that move >>>>a PV move. >>> >>>What do you do then? >> >>OK.. let's take the root. The first move is searched. I then use the PVS >>null-window on the remainder of the ply=1 moves. If one fails high on the >>null-window search, I re-search with the correct PVS beta value. If this >>search fails low, I pretend the fail-high didn't happen and keep right on >>searching without replacing the PV move. > >I don't understand. Isn't that the normal way of doing things? I mean, you >search with a null-window. You get a fail-high, that is the score is larger than >alpha+1. Then you research with the full window, as always in PVS. Then the >result of THIS search is the on that counts, right? So if this one fails low, >then why would you ever replace the PV move? What do you do if the null-window search fails high and then you run out of time? Normally a fail-high causes a PV change. I special-cased this to not happen on the null-window search unless the relaxed beta bound search confirms that the fail-high was real. > >>If it fails high on the re-search, >>It does become a PV move even if it then fails low on the 4th re-search... >> > >Again, isn't this just the way PVS is? > >>The bad case is failing high on the null-window search then failing low on >>the first re-search... that can make you play lemon moves... > >I don't understand this at all.. :) Most implementations cause _any_ fail-high to place that move first and make it a PV move, instantly. That can cause problems here...
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.