Kernel density estimate

[VTwin]

I'm looking for 2D kernel density estimation (KDE) code to contour data, but am having no luck finding pascal code.

I'd be grateful if anyone can share a pointer to existing code.

VTwin

[winni]

Hi!

No, nothing in fpc and there seems to be nothing in Turbo Pascal.

In the net you'll find sources for Python, C++, R and mathlab.
But not even pseudocode.

Sorry.

Winni

[VTwin]

Yes, there are a lot of implementations. The MATLAB, R, C++, and Python code I have found have a lot of dependencies and advanced options. I am content to start with a simple "rule of thumb" implementation (Silverman, 1986). I imagine I can figure it out, but was hoping there might be some existing code.

[jamie]

https://en.wikipedia.org/wiki/Kernel_density_estimation
If you look down below there is code....

Is that what you are looking for ?

[VTwin]

Yes, that is a good summary of the 1D KDE, although I need a 2D KDE. That was pretty much where I started looking, and I did not find any links there that seemed promising.

I can dig further into some of those links, but the MATLAB and R code is not straightforward (for me) to translate.

ESRI has a decent explanation of their implementation that I could probably code, but it uses a quartic kernel rather than the more common Gaussian kernel. I'm no expert, and just looking for a basic implementation.

[MathMan]

Better Wikipedia link is https://en.wikipedia.org/wiki/Multivariate_kernel_density_estimation I think - it also has a link to a a collection of peer-reviewed articles on multivariate KDE math.

If you are not shy of tackling C++ sources then libagf.sf.net (not checked if it still exists) might be a place to look at.

Another one is available here - https://gitlab.com/stochastic-control/StOpt (not checked if it still exists). It relates to arXiv:1712.00993v2 (from Oct 2018, so quite recent at least) which explicitely covers univariate, bivariate and multivariate approaches.

My search also didn't reveal a lot of links to sources and unfortunately none to Pascal.

Cheers,
MathMan

[VTwin]

Thanks MathMan. I did overlook that page, and it is helpful. I'm generally not shy about using C++ sources, I used it exclusively for a few years, although it is easy to write obtuse code.

libagf is still available on sourceforge with some documentation. I'll see what I can figure out.

Cheers,
VTwin