Re: [Bug-gnubg] RE: gnubg bearoff database query

[Joern Thyssen]

On Thu, Aug 28, 2003 at 04:09:28PM -0300, Albert Silver wrote
> (I'm forwarding this as someone in the list might be able to provide
> more information on the subject of database recommendations)
> 
> ------------------------------------------------------------
> 
> > Albert:        Could you advise me on which bearoff databases to
> generate
> > and why?  Although I find the
> > documentation in the "doc" folder to be good, conceptually I am a
> little
> > out
> > of my league.
> > My guess is that gnubg doesn't have to crunch as often if it can
> reference
> > these databases.  Is that right?  
> 
> Well, GNU doesn't really suffer from the classic bot problem of
> extremely poor bearin in non-contact positions thanks to its very good
> Race Net. The biggest advantage of a large database is that even those
> minute errors are improved thanks to a perfect database, and it doesn't
> run them through the race neural net, so it could be faster. 

Actually it's slower!

> > Also,  given that computing time is not
> > a
> > problem should I go with the two-sided bearoff databases rather than
> the
> > one-sided, and if so, should I generate as many that are reasonable?  
> 
> If you want bearin perfection, I'd build a one-sided 13-point database
> extending to the midpoint. This is the classic point of departure for
> non-contact bearins. The two-sided bases are for bearOFF positions as a
> rule (AFAIK), else the base would probably be too big. Feel free to
> experiment though, or query the mailing list. Would it be ok if I
> forwarded this? I'm fairly certain you'd get better answers on this
> subject that way.

The two-sided database tend to grow very fast with the number of points
and chequers. I've got Hugh Sconyer's full bearoff databases for 15
chequers vs. 15 chequers on 6 points, and it takes up 11 DVDs for a
total of 40+ GB for the close to 3 billion positions. 

The number of positions in a two-sided bearoff database for 6 chequers
on 13 points is more than 500,000,000,000,000,000 and would require 4
exabytes of storage. You just need 33 million new 120 GB harddisks :-)

For the corresponding one-sided databasre you only need 5 GB of storage.
BTW, I don't think that anyone has managed to generate such as database
for gnubg. I think MOrten Wang mananged to generated the database up to
the 12point.

Jørn

pgpgtHfwtelsJ.pgp
Description: PGP signature