Page 1 of 1

problem in the function model.auto

PostPosted: Mon Jun 06, 2016 2:27 pm
by Maxime Beauchamp
Hi,

I have some troubles when fitting some variograms.
For instance, if I use model.auto(vario,struct=c('Nugget Effect','Exponential','Spherical'),draw=F), the program seems to crash but if I specify the structures in an other order, let's say struct=c('Nugget Effect','Spherical','Exponential'), now it works.

The problem appears quite often and it's not always the same order of the argument struct that creates the bug.

Do you know where the problem comes from and how to solve it?

Best regards,

Maxime

Re: problem in the function model.auto

PostPosted: Mon Jun 20, 2016 9:27 pm
by Didier Renard
I was not able to reproduce the problem.
Could you please:
- indicate the version number
- possible download the latest one and check if the bug is still there
- optimally send the data set and the script

If not possible, could you add the verbose option to model.auto.
Another interesting test would be to switch ON the debugging option by typing:
debug.define("converge",1)
before submitting model.auto again.

Thanks for your help.

Re: problem in the function model.auto

PostPosted: Tue Jun 21, 2016 9:07 am
by Maxime Beauchamp
Hi,

Well in fact it seems I have identified the problem.
The bug happens when trying to fit a variogram in metric coordinates over a large domaine (Europe for instance). The distance between the points could thus be very important, up to a million meters.
If I artificially divide the coordinates by a factor of 1000 to have a pseudo kilometric coordinates system, then the problem never appears.

It seems that the procedure have some issues when dealing with very large distances.
Do you have an idea of what is the problem so ? If not, I can send you an example of the data and precise you the version of RGeostats used but I think it was the last one available on the site.

Thanks for your help,

Maxime