Re: [Bug-gnubg] Predefined move filters

[Gary Wong]

On Mon, Dec 30, 2002 at 08:49:35PM +0000, Joern Thyssen wrote:
> Correct. Using 1-ply may often remove the correct move, so that's why we
> generally skip pruning at odd plies.
> 
> The old setup has a setting "Skip pruning at 1-ply" or something similar
> for the same reason.

I'm not really convinved about this.  The old "skip 1-ply pruning" did
improve evaluations, at the cost of being slower -- if you turned it
on, instead of getting (say) all moves eval'ed at 0-ply, and the best
16 of those at 1-ply, and the best 8 of those at 2-ply, you got all at
0-ply and the best 16 at 2-ply.  This is definitely better, but
roughly twice as slow.

I think that a better choice would have been to turn "skip 1-ply
pruning" _off_, and increase the number of candidates.  That way, you
might get everything at 0-ply, the best 32 at 1-ply, and the best 16
of those at 2-ply -- I believe this is almost as fast as the choice of
filter above, and slightly stronger.

I think that including odd-ply evaluation in the filtering is a good
thing, as long as the parameters are set intelligently.  (I have no
idea what the most intelligent set of parameters are, of course :-)

Cheers,
Gary.
-- 
   Gary Wong           address@hidden           http://www.cs.arizona.edu/~gary/