Re: octcdf questions

[Alexander Barth]

Hi Joan

On Wed, May 29, 2013 at 12:42 PM, Joan Pau Beltran <address@hidden> wrote:

Hi Alexander,

Thank you for your response, now it is clear to me how autonan and autoscale work.
Probably a comment in the documentation's example would help here.

I added an example.

BTW which are the actual names: ncautonan and ncautoscale or autonan and autoscale?

both should work.
 

The later are undefined in my version of octave with ocatve-octcdf.

Regarding the other failure, I realized it only happens to me with variables with several dimensions (e.g matrices).
This probably points out to a misuse. Let me reword my example (forgetting autonan and autoscale issue):

nc = netcdf('myncfile.nc','c');

nc('dim1') = 3;

nc('dim2') = 2;

nc{'var1'} = ncdouble('dim1');

nc{'var2'} = ncdouble('dim1', 'dim2');
nc{'var1'}(:)   = rand(3,1); % this works.
nc{'var1'}(1:3) = rand(3,1); % this works, too.
nc{'var2'}(:)   = rand(3,2); % this works.
nc{'var2'}(1:6) = rand(3,2); % this fails: octcdf: unknown index specification: type <unknown_type>
nc{'var2'}(:,:)     = rand(3,2); % this works.
nc{'var2'}(1:3,1:2) = rand(3,2); % this works, too.

So does it mean that I should explicitly indicate the slice along EACH DIMENSION?

Yes! unless you use (:) which is a special case.

 

Another possibility is to use either a single magic colon, or a magic colon for each dimension. Will the last option work with variables with record dimensions?

no

Alex

Thanks again!

2013/5/28 Alexander Barth <address@hidden>

Hi Joan

For autonan and autoscale you should do something like this:

What is strange is that both assignments work for me using octcdf 1.1.6 and octave 3.6.2

nv(:) = 1:5; % to use the autonan and autoscale
nv(1:5) = 1:5;

or in this form:

nc{'var1'}(1:5) = 1:5; % without autonan and autoscale

What does happen when you try "nv(1:5) = 1:5;" ?

Can you run nctest?

Joan Pau Beltran