I personally would not spend time on this minor issue at this point. If you
want to save space you can try and train dedicated nets, which might
outperform the one-sided approach.

-Joseph

On Wed, 25 Sep 2019 at 08:37, Philippe Michel <[email protected]>
wrote:

>
> On Sat, Sep 21, 2019 at 07:15:05PM -0700, Theodore Hwa wrote:
>
> > Any comments, suggestions, contributions are appreciated!
>
> In the comments on the github page you write:
>
> Note that due to its size, the two-sided bearoff database gnubg_ts0.bd
> is currently not packaged.
>
> Maybe you could package a smaller two-sided database ? The default one
> with 6 checkers is 7M but one with only 4 checkers is much smaller
> (350k). You can create it with "makebearoff -t 6x4 -f gnubg_ts0.bd".
>
> If I remember correctly, for some time Russ Allbery had been reluctant
> to install it in the Debian port (but the issue was the time that may be
> needed to compute it at installation time, not its size per se).
>
> I suppose that, when these databases were introduced, it has been
> studied how the two-sided one improved accuracy over the one-sided one
> for various numbers of checkers left. Does anyone has these results ?
>
> _______________________________________________
> Bug-gnubg mailing list
> [email protected]
> https://lists.gnu.org/mailman/listinfo/bug-gnubg
>
_______________________________________________
Bug-gnubg mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/bug-gnubg

Reply via email to