Re: Test data for (m-code) semiovariogram?

[fork]

PhilipNienhuis <pr.nienhuis <at> hccnet.nl> writes:

> 
> forkandwait wrote

(SNIP)

> > Does anyone know of a good test dataset, with known correct input and
> > ouptut, 
> > that I can run to verify the code?
> 
> Just a little bit of googling turned up this:
> 
> http://www.ats.ucla.edu/stat/mult_pkg/faq/general/spatial_autocorr.htm
> &
> http://www.ats.ucla.edu/stat/sas/faq/SAS_variogram_fit.htm

Cool.  I did google I promise!

> The grid size isn't specified but I'd suggest to just assume 100x100 m (or
> feet, or miles, or whatever.... if only the variogram shape makes sense)
> 
> I agree that it was harder than I initially thought to uncover easily,
> immediately usable test data. If it gets too hard for you I'll try to dig
> deep into my own archives (floppies, gnah). No guarantees though.

Lets see how it goes...  If it is easy, I would be interested...

> At the same time, while googling for test data, I almost got drowned in
> ready-baked variogram procedures - indeed, several from Matlab Central ;-P

I found some Matlab code, but the ones' that looked solid also required 
setting globals or dealing with MEX or some other stuff I didn't like in my 
search for simplicity and portability.
 
> The statistics package may be a good target; I'd suggest to add a new
> category "geostatistics"

I imagine just writing a single function -- semivariogram() with an option 
for returning a fitted model -- and to start a whole new package seems silly 
for that.  Even with 2 funcs -- kriging() -- it seems silly.  Plus I only 
probably have time to submit a patch at the moment, not setup and maintain a 
package.

Thanks for the info!