Author: Bas Hamstra
Date: 02:43:38 09/29/99
Go up one level in this thread
On September 29, 1999 at 01:20:45, Gian-Carlo Pascutto wrote: >On September 28, 1999 at 15:43:43, Robert Hyatt wrote: > >>On September 28, 1999 at 13:42:00, Gian-Carlo Pascutto wrote: >> >>> >>>Also, does the actual depth the program reaches play part in >>>this consideration ? (e.g. if you only get 5 ply then R=2 or >>>R=3 are out of the question?) >>> >> >>they are _very_ dangerous at that depth.. because any null-move will >>take you right into the q-search, which is pretty simple-minded. It works >>far better at deeper depths... > >Ok so what about deciding the R factor by looking up how many plies >there are still below the null-move search ? i.e. trying to ensure >there is always a non-null move search in the null-move search (ahem) >This seems workable as it always catches at least the simple one-move >threats. > >Is something like R=1 doable for 4-5 ply searches ? At what depth can >one consider taking it to R=2 ? > >-- >GCP r=2 is pretty standard for *whatever* depth. I would say: just set r=2 and be sure your engine thinks deep enough. If you set r=2 on decent hardware, you are *not* thinking 5 ply, but 10+. Or else you have something other than nullmove to worry about. With r=2 you think at least 2 plies deeper than without null... In case you play blitz/bullet on slow hardware: If Scrappy (=Crafty on sometimes very slow hardware) thinks 5 ply, it just uses r=2. If you use r=3 or more near the leafs it can cause search instabilities, and you sometimes do *more* nodes...
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.