On Sunday 11 May 2008 20:58, Ethan A Merritt wrote:
> On Sunday 11 May 2008 20:40, Philipp K. Janert wrote:
> > I have dusted off patch 1827826 (more smoothing
> > algorithms for dgrid3d) and made it to work with the
> > current development version.
> >
> > Are there any objections to inclusion into the main
> > branch at this time?
>
> Could you summarize for us what tests you have run to confirm
> that it functions as advertised?
I have run the complete gnuplot unit-test suite
against it. ;-)
I have taken some sample data set and tried
the new algorithms out on them and visually
inspected the results. The "gnu-valley" mini
data set from the gnuplot documentation is
pretty informative.
>
> Does it gracefully handle missing or non-uniform data?
Yes. Both of these are features that dgrid3d itself provides,
and which I have not modified. I merely have added some
additional smoothing kernels (and syntax and options to
manage them).
>
> Should it come with guidance or warnings about which function
> is suitable for what data?
I don't think this is really necessary. The Gaussian
kernel is probably the most general, all-purpose
kernel in this patch. I don't want to patronize the
reader, but we can include a sentence like "Unless
you have special needs, the Gaussian kernel is
likely to give satisfactory results."
>
> Maybe a demo like the one on the Wikipedia page at
> http://en.wikipedia.org/wiki/Window_function
I can provide some demos for inclusion into the demo/
folder. I think that is a good idea.
Best,
Ph.
|