topic Re: Cubic spline fitting options need improvement in Intel® oneAPI Math Kernel Library

Cubic spline fitting options need improvement

Andrew_Smith — Fri, 19 Jan 2024 10:34:35 GMT

We use shape preserving spline curves a lot but find the Akima offering in MKL has some shortcomings.

a) The ends of the spline are very flat. There are some modified Akima schemes which improve this aspect. One is offered in Boost for free and another in Alglib and other commercial packages.

b) The Akima spline does not have continuous second derivative. The one we use where we need this inserts additional breakpoints to preserve the shape (monotonicity).

Re: Cubic spline fitting options need improvement

Gennady_F_Intel — Wed, 24 Jan 2024 05:05:08 GMT

Thanks Andrew. We will add this topic to the future schedule.

Re: Cubic spline fitting options need improvement

Andrey_F_Intel — Fri, 11 Oct 2024 08:49:38 GMT

Hi Andrew. Sorry for the delay.

Regarding a). Could you please add more information about the use case?

It would be ideal if you shared the dataset (x, y, sites) when you see flat ends of the Akima spline (if possible).

The thing is that we don't see any issues with our datasets.

Re: Cubic spline fitting options need improvement

Andrey_F_Intel — Mon, 28 Oct 2024 10:45:40 GMT

@Andrew_Smith, please, let us know if the issue is still relevant

Re: Cubic spline fitting options need improvement

Andrew_Smith — Tue, 29 Oct 2024 17:33:22 GMT

I am happy to close this. I find the the MKL Akima spline with Not-A-Knot end option suites my needs. It models a radius as well as the Alglib Akima spline.