Author: Andrew Platt
Date: 13:09:24 09/25/04
Go up one level in this thread
On September 25, 2004 at 08:48:48, Harald Lüßen wrote: >>I don't know if this was from the posts I was making about problems I was having >>but that's not exactly the solution you want. The key point is that even when >>verifying a null move search you are still doing null moves lower down the tree >>and they could be causing problems too. However, you won't verify those becuase >>you are in verify mode. So now you are doing an R=3 null move and could miss >>some things. > >What mode? Are the nullmove algorithms in deeper plies dependent >on a decision of a shallow ply? I thought they are independent >with the exception of avoiding double nullmoves. A possible >deep ply nullmove is done when there are pieces and verified >when it fails high. In the original paper there is no difference between the null mode done when verifying and when not verifying. My speculation was that this was not completely accurate because if the whole point is to verify the null move was correct, you shouldn't use the same null move lower down where you don't get to verify. (Unless I missed something and you're supposed to reset the verify flag lower down the tree). >And when a search with a small depths with or without nullmove >backs up its score I could trust it whether I am in a normal >search or in a verification search. That is the trick with >recursion. Where am I wrong? Is it explained in the original >verification search document? There are only two scores you "trust" with a null move: A cutoff and a mate threat. Since those never end up in the PV you just have to accept them. Andy.
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.