Hello,
Is there an implementation of multidimensional interpolation done in Numba? I have not found one that can be used with njit.
Best
It’s well maintained and fast.
Hi. I have a similar question. From what I have seen, the splines in EconForge rely on tensor product grids. Is there a multidimensional interpolation scheme consistent with the use of quasi-random grids? Essentially, I am looking for something in which the number of grid points does not explode with the number of dimensions (state variables).
Thanks!
Hi Mario,
I’m maintaining the Econforge/interpolation library. The only methods we have right now for higher dimensions are complete polynomials and smolyak collocations, but we could consider adding more in the future. I’m all ear if you know a methods, which can alleviate the curse of dimensionality. I think we all are 
What are you thinking about?
If it is just about fitting the interpolation coefficients through a regression rather than exact collocation, that could easily be added to the library.
At some point there was also a WIP implementation of basis matrices, with jitted vectorized computations. But the basis objects themselves were not part where not jitted objects themselves, which is why it got temporarily removed. I could point you to it, with no guarantee, if you think it can be useful to you.
Then there is the whole world of nonlinear interpolator: RBF and co., neural networks. Although, I haven’t worked on it, I would say a simple implementation of RBF in numba would be feasible and desirable. N.N. requires a completely different architecture (tensorflow is your friend).
Hope this helps. I’m afraid there is no definitive answer.
Best,
Pablo
1 Like