Re: [Bug-gnubg] Questions about gnubg-nn tools

[Massimiliano Maini]

On 5 January 2012 02:20, Joseph Heled <address@hidden> wrote:
> Seems like you got a very very slightly better race net, but I would be
> surprised if it makes a difference in real life.
>
> Would be much more interesting to
>   - get a better contact or crashed net
>   - expand the roll-out database for all categories (should be easy with the
> current availability of cycles)
>   - improve cube decisions (this is a hard one)
>   - improve back game evaluation and play (very hard one)
>
> -Joseph

One thing that has always puzzled me is the strange behavior of the strength
of gnubg at different plies. The last large scale study (done to compare the
existing bots to the new extreme gammon), is resumed here:

  http://www.extremegammon.com/studies.aspx

Checker play is fine, error goes down as plies go up.
However, for cube actions, the situation is very strange:

3ply does much better than 2ply and 4ply on missed doubles and wrong takes.
But 3ply does terribly worse than 2ply and 4ply on wrong doubles and
wrong passes.

It has always been accounted to the so called "odd-even" effect, but
no other bot
seems to be affected as much as gnubg.

MaX.