Re: [Bug-gnubg] Bug: 1-sided database

[Joern Thyssen]

On Wed, Sep 24, 2003 at 09:23:03AM -0400, Christopher D. Yep wrote
> At 01:19 PM 9/24/2003 +0000, Jørn Thyssen wrote:
> 
> >hmm, I get:
> >
> >26/36 + 10/36 * 2/36 = 956/1296 = 73.7654321% gammons
> >
> >Which is the same as the database. Did you notice that 2-1 loses gammon
> >for O?
> 
> No, I didn't notice that 2-1 misses.  Sorry!! I must be half-asleep this 
> morning (in the U.S.).

No problem.

As an interesting side note I was actually in doubt how to handle
gammons in the one sided database. 

When the database is generated the best moves are found by selecting the
move with the lowest average number of rolls to bear off. But what about
the gammon distribution? 

As I've implemented it the bearoff and gammon dist are totally
independent, so the bearoff distribution for a given position may be
based on different "best moves" than the gammon distribution. In
principle this could lead to wrong cubeless gwcs and gammons for
positions where gammon is possible: If I'm playing to save gammon I'll
select my moves primarily based on the gammon distribution so the
bearoff distribution is "wrong" since it may be based on different
moves, hence the calculated gwc is wrong. If I'm playing to win I'll
select my moves based on the bearoff distribution, so the gammon dist is
"wrong", hence the calculated gammon percentage is wrong.

Fortunately, I'm saved by the fact that when gammons are possible the
gwc is typically zero and vice versa: when wins are possible gammons are
usually not. However, I've never verified this...

Jørn

pgpLk0HJHthwd.pgp
Description: PGP signature