Author: Robert Hyatt
Date: 07:09:52 12/28/99
Go up one level in this thread
On December 28, 1999 at 04:15:21, Bas Hamstra wrote: >On December 27, 1999 at 10:24:30, Robert Hyatt wrote: > >>On December 27, 1999 at 05:14:47, Edward Heddle wrote: >> >>>I'm new here. I am not actually involved in programming, but more the theory of >>>programming. >>> >>>My question is; how do null move reductions work; i.e, at what points in the >>>search tree are null moves tried? >>> >>>Thanks >>> >>>Ed ;) >> >> >>Everywhere. Any position where you can let your opponent move twice in a >>row, and you still fail high, is obviously a very good position for you (or >>a very bad one for him.) > >Still it is strange that this worked bad at shallow depths. That same kind of >errors must still be in the nullmove tree. If > >I am still not entirely convinced that with nullmove is always better. If I let >my engine play GNU for example with R=2 and with nullmove off, it doesn't seem >to make a big difference at blitz. Null move hides things. If you do a 5 ply search, you make a move at ply 1, your opponent plays Qh6 at ply=2 (he has a pawn at f6, you have _nothing_ to defend g7 against Qg7 mate). Now if you try a null-move, R=2, you end up in the q-search where you won't see Qg7+ unless you include this kind of move in your capture search. So you don't see it, you make the move at ply=1 and instantly lose. The move at ply=1 might be taking what appears to be a free rook. But it costs _dearly_... At deeper depths, the errors occur within 3 plies of the tips, which is far from the root. Errors are less likely, although still possible to be sure..
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.