[SciPy-Dev] schedule for scipy 0.9.0 [scipy.interpolate]

Charles R Harris charlesr.harris@gmail....
Fri Nov 12 11:25:45 CST 2010


On Fri, Nov 12, 2010 at 9:48 AM, <josef.pktd@gmail.com> wrote:

> On Fri, Nov 12, 2010 at 11:21 AM, Pauli Virtanen <pav@iki.fi> wrote:
> > Fri, 12 Nov 2010 23:45:39 +0800, Ralf Gommers wrote:
> >> I'm interested to hear what you think about releasing scipy 0.9.0 in the
> >> near future, and if you have any things that should go in or be fixed
> >> for 0.9.0. If we want to keep to a ~6 month release schedule then we
> >> should branch around the end of this month and have a final release in
> >> January.
> >
> > Sounds OK to me. I think we should try to time it so that it gets in
> > Ubuntu.
> >
> > Some tasks that would be nice to do before that in scipy.interpolate:
> >
> >  - Write a interpolation routine for N-d *structured* interpolation.
> >    (ndimage.map_coordinates is difficult to find, so it would be useful
> >    to have an interpolation-oriented routine for that in
> >    scipy.interpolate).
> >
> >  - Deprecate interp2d, as it does not work well for most problems,
> >    and its calling convention is a bit wonky. People should use either
> >    `griddata` or splines depending on what they want to do.
> >
> >  - Clean up the spline routines:
> >
> >    - We have two incompatible implementations for 1-D splines there,
> >      on from FITPACK, and a pure-Python one by Travis. I don't remember
> >      if one of them has an advantage over the other.
> >
> >      I think that either `splmake` the one should be deprecated, or
> >      the spline representation it uses made compatible with FITPACK.
> >
> >    - Deprecate the bispl* and spl* routines, leaving only the
> >      object-oriented interface.
>
> These are in some cases more robust and give better control, the
> classes look a bit "messy". I would only depreciate them if the
> classes get some makeover.
>

Which classes? If the ones in interpolate, I agree that they are a bit
limited.

The simple case of running a 1-d spline through a set of points, with some
provision for handling the boundaty conditions, is often  done without using
b-splines. Doing tricky things with varying degrees of discontinuity at the
knot points or generating least squares fits is better done with b-splines.
If we are going to use just one package I would stick with fitpack, or an
equivalent subset of opennurbs. The latter is a large package in c++
intended for CAD with the usual tangle of inter-dependencies, but the core
bits don't look like too much trouble to disintangle and convert to simple
C.

Chuck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.scipy.org/pipermail/scipy-dev/attachments/20101112/ac34563b/attachment.html 


More information about the SciPy-Dev mailing list