coefficients of spline interpolation in scipy

matlab, python, scipy

Solution

I'm not sure there is any way to get exactly those coefficients from scipy. What `scipy.interpolate.splrep` gives you is the coefficients for the knots for a b-spline. What Matlab's spline gives you appears to be the partial polynomial coefficients describing the cubic equations connecting the points you pass in, which leads me to believe that the Matlab spline is a control-point based spline such as a Hermite or Catmull-Rom instead of a b-spline.

However, `scipy.interpolate.interpolate.spltopp` does provide a way to get the partial polynomial coefficients of a b-spline. Unfortunately, it doesn't seem to work very well.

>>> import scipy.interpolate
>>> x = [0, 1, 2, 3]
>>> y = [0, 1, 4, 0]
>>> tck = scipy.interpolate.splrep(x, y)
>>> tck
Out: 
    (array([ 0.,  0.,  0.,  0.,  3.,  3.,  3.,  3.]),
    array([  3.19142761e-16,  -3.00000000e+00,   1.05000000e+01,
        0.00000000e+00,   0.00000000e+00,   0.00000000e+00,
        0.00000000e+00,   0.00000000e+00]),
    3)

>>> pp = scipy.interpolate.interpolate.spltopp(tck[0][1:-1], tck[1], tck[2])

>>> pp.coeffs.T
Out: 
    array([[ -4.54540394e-322,   0.00000000e+000,   0.00000000e+000,
           0.00000000e+000],
        [ -4.54540394e-322,   0.00000000e+000,   0.00000000e+000,
           0.00000000e+000],
        [ -4.54540394e-322,   0.00000000e+000,   0.00000000e+000,
           0.00000000e+000],
        [  0.00000000e+000,   0.00000000e+000,   0.00000000e+000,
           0.00000000e+000],
        [  0.00000000e+000,   0.00000000e+000,   0.00000000e+000,
           0.00000000e+000]])

Note that there is one set of coefficients per knot, not one for each of the original points passed in. Also, multiplying the coefficients by the b-spline basis matrix doesn't seem to be very helpful.

>>> bsbm = array([[-1,  3, -3,  1], [ 3, -6,  3,  0], [-3,  0,  3,  0], 
                 [ 1,  4,  1,  0]]) * 1.0/6
Out: 
    array([[-0.16666667,  0.5       , -0.5       ,  0.16666667],
        [ 0.5       , -1.        ,  0.5       ,  0.        ],
        [-0.5       ,  0.        ,  0.5       ,  0.        ],
        [ 0.16666667,  0.66666667,  0.16666667,  0.        ]])

>>> dot(pp.coeffs.T, bsbm)
Out: 
    array([[  7.41098469e-323,  -2.27270197e-322,   2.27270197e-322,
           -7.41098469e-323],
        [  7.41098469e-323,  -2.27270197e-322,   2.27270197e-322,
           -7.41098469e-323],
        [  7.41098469e-323,  -2.27270197e-322,   2.27270197e-322,
           -7.41098469e-323],
        [  0.00000000e+000,   0.00000000e+000,   0.00000000e+000,
           0.00000000e+000],
        [  0.00000000e+000,   0.00000000e+000,   0.00000000e+000,
           0.00000000e+000]])

The FORTRAN Piecewise Polynomial Package, PPPack, has a command `bsplpp` that converts from B-spline to piecewise polynomial form, which may serve your needs. Unfortunately, there isn't a Python wrapper for PPPack at this time.

Problem

I want to calculate the coefficients of a spline interpolation by scipy. In MATLAB: ``` x=[0:3]; y=[0,1,4,0]; spl=spline(x,y); disp(spl.coefs); ``` and it will return: ``` ans = -1.5000 5.5000 -3.0000 0 -1.5000 1.0000 3.5000 1.0000 -1.5000 -3.5000 1.0000 4.0000 ``` But i can't do that by interpolate.splrep in scipy. Can you tell me how to calc it?

Original source