Example of `scipy.interpolate.BSpline.design_matrix` equivalent with`skfda.representation.basis.BSplineBasis` about scikit-fda

[vnmabus]

The BSplineBasis object in scikit-fda is not intended to be a replacement of SciPy's B-splines (in fact, they are used internally). This object is just used to define a particular basis of functions for a FDataBasis object. Although some procedures, such as the computation of the design matrix, can be done with both, there exist some differences and I am not sure if I would recommend using scikit-fda for that purpose. First, as SciPy's B-splines are used internally, the scikit-fda implementation can only be slower. Secondly, the implementation of design_matrix in SciPy returns a sparse matrix, which will be more efficient in both memory and speed for a high number of basis elements.

Thus, unless you can provide a better explanation of why that should be done, I do not think that this is useful for most users.

Finally, note that there is an easier and faster way to compute that design matrix in scikit-fda:

fda_matrix = fda_basis(x)[..., 0]

[avivajpeyi]

Ah I see! Thanks @vnmabus. In that case maybe we can close this issue.

Tbh the main reason I am using scikit-fda now instead of scipy' Bsplines is because I need to generate a penalty matrix for my basis (which I didn't know how to do with scipy's Bsplines, but can do with scikit-fda)

This might be another issue, but since the scipy design matrix is more efficient, is it possible to generate a penalty matrix for that via the scikit-fda interface?

[vnmabus]

Unfortunately no. However for a small number of basis elements the difference in performance is probably not that much.

I will close this issue as not planned, then.