Author: Dieter Buerssner
Date: 10:45:23 06/07/05
Go up one level in this thread
On June 06, 2005 at 22:49:52, Robert Hyatt wrote:
>There's nothing wrong with any of that, except that most use a null-window
>search today, which means that 99.9999% of the positions searched have
>alpha=beta-1, so that adjusting the bounds can't actually happen, as if you were
>to be able to adjust the lower bound, you already have enough info to produce a
>fail high, and vice versa...
The 99.9999% first looked like an exaggeration to me (it is one, for normal
engines doing say a not too long search, but I won't discuss how many digits 9
there really should be - the intend of the message is clear). So, I thought
count the nodes (I did not count qsearch nodes). I always use the zero window
from the second move onwards here (no other preconditions). For a depth 12
search of the start position I got:
pv nodes 516, non pv nodes 5390768 99.9904% researches 246
This was about one minute (most nodes are not shown, because they are in
qsearch, where I don't use the PVs algorithm, either). researches is the number
of times, the engine had to do a research inside the tree, because score was >
alpha and < beta, and the first search was done with zero window. Doesn't sound
516 pv nodes too small? Can anybody try with his engine? I just added
#if COUNT_PV_NODES
if (alpha == beta-1)
non_pv_nodes++;
else
pv_nodes++;
#endif
at the start of the absearch function, and some obvious initialization and
printing code.
Regards,
Dieter
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.