How to compute a spline value from a set of knots and coefficents given by Scipys SmoothBivariateSpline

Question

I am using python to fit a spline to my data. I then need to be able export the set of knots and coefficients to a C program for regular use. I am using scipy.interpolate.SmoothBivariateSpline to fit the spline and use the methods get_coeffs() and get_knots() to have a copy of the knots and coefficients.

Ultimately I need to rewrite the evaluation method (ev()) so that I can use my spline in a C environment. Can anyone guide me to some resources on how to write this method?

Answer 1

Looking through the code, evaluations of bivariate FITPACK based splines delegate to dfitpack.bispev, https://github.com/scipy/scipy/blob/master/scipy/interpolate/fitpack2.py#L844

The latter function is an f2py wrapper, https://github.com/scipy/scipy/blob/master/scipy/interpolate/src/fitpack.pyf#L352, of a Fortran routine of the same name, https://github.com/scipy/scipy/blob/master/scipy/interpolate/fitpack/bispev.f, which invokes https://github.com/scipy/scipy/blob/master/scipy/interpolate/fitpack/fpbisp.f

One gotcha to be careful with is exactly how the end knots are handled by get_knots and get_coefs, https://github.com/scipy/scipy/blob/master/scipy/interpolate/fitpack2.py#L780

This all is probably discussed in all the gory detail in the Dierckx's book. Or maybe not --- the computational routine looks like a fairly straightforward generalization of the de Boor's scheme for 2D. Devil's in the details though.