Speed...

[dncnwtts]

Currently the speed of calling clee-fast is on the order of calling CAMB directly. This is fine if you don't want to use CAMB itself, but the speed increase of 2-5 that I've observed when going from CAMB to clee-fast is untenable for actual likelihood evaluations, especially when for the matrix inversions we are using, the limiting time scale is O(10 ms).

There are a couple of options for interpolation, one is "kriging", there's also simply fitting a polynomial of some order at each point. I would think that scikit-learn has some module for dealing with this exact problem.