Re: Using the STk package for Kriging

[Julien Bect]

Le 11/11/2016 à 14:02, Julien Bect a écrit :
> Le 11/11/2016 à 12:33, Francesco Potortì a écrit :
> > I am experimenting with Kriging interpolation and I am studying the STK
> > package.
> >
> > I am surprised at the fact that the provided covariance functions are
> > Metérn, Gauss, and empirical.  The commonly used exponential and
> > spherical models are missing, like other common models.
> >
> > Sure, I can write my own covariance function, but since I am a beginner
> > at Kriging I wonder why exponential and spherical are missing: is simply
> > lack of manpower or is there a reason why those functions are not
> > implemented?  Or maybe they are there under a different name or the 
> > like?
> >
> > Specifically, I need a spherical variogram: if it is not there, are
> > there people here that have written one for themselves and would share
> > it?
>
> Hi Francesco,
>
> You're right, the number of available covariance functions in STK is 
> currently quite limited.
>
> The main reason is that the way covariance functions are written in 
> STK makes it difficult to expand the catalog of available covariance 
> functions properly without duplicating a lot of code (code that 
> handles a computation cache).  Future evolutions of STK will make it 
> much easier.  Which is why I was wiaiting for this evolution to happen 
> before adding more covariance functions... The main reason why this 
> hasn't happened yet is lack of manpower, as you said. Recent 
> developments in STK were simply heading in a different direction ;-)
>
> Now, to answer your request more specifically :
>
> The "exponential variogram" is actually the same thing as the Matern 
> 1/2 covariance function, so it is available as a special case of the 
> Matern covariance function (if you're interested in parameter 
> estimation, however, you will have to make a wrapper that "fixes" the 
> shape parameter to 1/2).
>
> The spherical covariance is not available yet, but I will add it soon, 
> since you need it ;-)

Hi Francesco,

I have added the exponential and spherical covariance functions to STK 
(see changeset [1]).

It will be available in STK 2.4.0, which I hope should be released 
soon... (within a month or two).

In the meantime, you can try the beta version that is available from 
SourceForge [2].

The functions that you are looking for are : stk_expcov_*, stk_sphcov_*, 
stk_rbf_exponential, stk_rbf_spherical.

Please keep in mind that this is just a beta, and report any problem 
that you encounter.

@++
Julien

[1] 
https://sourceforge.net/p/octave/stk/ci/49f4ea0206a6aa2407e36294fb4fec7c91902a4b/

[2] https://sourceforge.net/projects/kriging/files/stk/