Hello devs,
I have a situation where almost 90% of my code is numerical computation, with 10% for reading data in the beginning. Numba has really helped me to reduce the computation time while maintaining excellent readability.
I have to use interpolation in my code in between for some data. I use scipy’s ‘splrep’ to calculate coefficients at the beginning and further ‘splev’ inside the program for interpolation. The problem is that the subroutine containing ‘splev’ can’t be jitted. So some python overhead comes in between my numerical computation because of just one subroutine which has to call the interpolation subroutine. What is the best course of action in this situation, (1) Should I think about extending numba to recogonise ‘splev’ or (2) Should I just write a python version of ‘splev’ and jit the same. If someone can provide an example for option (1) that is posted elsewhere it would be very helpful.
Thank you