[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: 960523: A Character String Dimension

>Organization: Institute for Stratospheric Chemistry(ICG-1), Institute for the 
>Chemistry and Dynamics of the Geosphere
>Keywords: 199605231111.AA16583

Danny,

> If I am most specific hopfillxy it should be clearer.
> 
> What I would like to do is have a data structure that has at least two 
> dimentions. An unlimited time dimension and a parameter dimension that could 
> typically be between 50 and 300 in length. I would like to assign
> each parameter with a dimension value that is a string i.e. H2O or  HNO3
> or PV etc.

Currently, netCDF only supports one variable-length dimension per file, the
unlimited dimension.  When another variable-length dimension is needed, e.g
for "ragged arrays" with variable row-length, there are several approaches
to simulating it:

  - Just use a maximum dimension (e.g. 300) for the rows and waste space
    for those variables needing less than the full 300.  This can be
    unacceptable if the maximum dimension is much larger than the typical
    value, since it wastes so much space.

  - Instead of a two-dimensional variable, use two one-dimensional
    variables, one of which is an index into the other, that tells where
    each variable-length row should start.  This saves space, but is
    inefficient if you later have to extend any rows or if the rows change
    sizes frequently, requiring some sort of garbage collection.

  - Use a different fixed dimension for each variable-sized row.  As you
    have noted, this runs up against the arbitrary limit of 100 dimensions
    per file, and requires copying all the data if any of the row sizes ever
    change. 

  - Use multiple files for multiple unlimited dimensions.  This won't work
    if a single variable needs two variable dimensions.

> My desire to do this was because I thought that if I had to define a dimension
> for each parameter then I would quickly run foul of the limit(albeit 
> arbitrary) of 100 dimension. However I have now realised that what I can 
> easily do is just define each parameter to a variable that has the requisit 
> string for its name.  The only limitation that this gives me is that it is 
> somewhat inelegent
> and will be a little slow to place data into the data structure. So if there 
> is some
> obvious way for me to achieve this please let me know however if it requires 
> any thought on your part then  I think I have a method that will be adequate 
> so please do not go to too much  trouble.
    
That's another solution I hadn't thought of.  I don't know any other
obvious solutions with the current format.

--Russ

______________________________________________________________________________

Russ Rew                                           UCAR Unidata Program
address@hidden                              http://www.unidata.ucar.edu