[SciPy-dev] Status of sandbox.spline and my other code.

John Travers jtravs@gmail....
Tue Apr 8 03:56:24 CDT 2008


---------- Forwarded message ----------
From: John Travers <jtravs@gmail.com>
Date: Tue, Apr 8, 2008 at 9:55 AM
Subject: Re: [SciPy-dev] Status of sandbox.spline and my other code.
To: Jarrod Millman <millman@berkeley.edu>


On Mon, Apr 7, 2008 at 11:56 AM, Jarrod Millman <millman@berkeley.edu> wrote:
 >
 >  Hey John,
 >
 >  I am hosting a NumPy/SciPy sprint this week at UC Berkeley and among
 >  other things am hoping to make some progress on cleaning up the
 >  various interpolation/spline code in SciPy as well as continue
 >  removing the scipy sandbox.  Do you have any more plans for your
 >  sandbox.spline code?  Any thoughts on where it should go or what still
 >  needs to be done?
 >

 The code in sandbox.spline should be a complete replacement for the
 code in scipy.interpolate.fitpack and fitpack2 including the f2py
 wrappers of the fitpack fortran code (i.e. replacing the hand crafted
 c wrappers). Apart from using only f2py, a number of improvements were
 made to the docstrings and error checking and some bugs fixed and
 quite a few unit tests added.

 Originally I was advocating that we have a separate spline package
 with the fitpack stuff and a separate interpolate package that can
 utilise either the spline or rbf of delaunay etc. packages as these
 can all be used for more than just interpolation. This didn't seem to
 gain any traction on the mailing list, so I left it. I would still
 advocate this, but recognise the reasons that people might not want to
 expand the number of packages so much.

 I left this code about a year ago and then got bogged down with my
 thesis work which I finished a few weeks back. If they
 code/improvements might be used then I can find the time this week to
 check it is all working and still up to date. Then I could either
 supply a set of patches against the main interpolate code which could
 be more easily understood, or provide a mercurial/bzr branch for
 someone else to merge from (if the bzr/mercurial stuff is setup yet).

 Let me know if that might be useful. There are also further things
 that were on my todo list for interpolate, including using
 RectBivariateSPline where possible as it is much better for
 interpolation than the currently used surfit. (I have a unit test
 which segfaults the current code, but not teh RectBivariateSpline
 code).

 Hope this helps,
 John


More information about the Scipy-dev mailing list