On Tue, Jun 02, 2026 at 08:41:06AM -0400, Nikolas Nyby wrote:
> Hello,
> Debian's gnubg package uses this patch to put the bearoff database in /var
> rather than /usr. I don't see why this patch can't be included in upstream
> gnubg, as /var does seem to be the right place to put this.
> 
> I wanted to open a discussion on this though, since I am not so familiar
> with how the bearoff database is used exactly.

There is an explanation about what was done in Debian (and why it is in 
fact not ideal) in this old message: 
https://lists.gnu.org/archive/html/bug-gnubg/2022-06/msg00010.html

There are two databases in fact:

- a two-sided database for very short bearoffs (a few chequers 
remaining, all in their respective inner boards), containing exact 
equities in money games. The racing neural net isn't expected to handle 
these positions properly. It is really important to have it.

- a one-sided database containing an histogram of the average number of 
moves needed to bear off one player's chequers. Comparing the histograms 
of both sides gives the cubeless winning chance for the side on roll. 
This is an improvement to the racing neural network for relatively short 
races.

The default databases are relatively small. The user can build larger 
ones. It would take a few hours for the largest supported sizes (less 
than 2 Gb).

Reply via email to