Re: [netcdfgroup] How to represent this data in NetCDF

• To: Dave Allured <dave.allured@xxxxxxxx>
• Subject: Re: [netcdfgroup] How to represent this data in NetCDF
• From: Jürgen Lorenz Simon <juergen_simon@xxxxxxx>
• Date: Tue, 31 Jul 2012 10:57:52 +0200

Hi,

On 30.07.2012, at 20:53, Dave Allured wrote:

> Define each "object" independently, except all objects have one
> dimension in common.  There will be no formal "array of objects", but
> you can access them automatically in a program, via an array of
> variable names.
> 
> This is a simplistic view of what I think you requested.  Will this work?
> 
> dimensions:
>  d = 30 ;
>  nx = 5 ;
>  ny = 7 ;
>  nz = 11 ;
> variables:
>  float x(nx,d) ;
>  float y(ny,d) ;
>  float z(nz,d) ;

I was trying to avoid the creation of one dimension per variable. Apart from 
that, I've done it just so, one variable per object, last dimension d is 
constant, N varies. 
Here is the code I used. It runs through without any exceptions but the error 
checking is also non-existent (yet). However, it does not create any variables. 
I'm going to add error-code checks today, but perhaps someone could tell me at 
first glance what went wrong:

        using std::stringstream;
        
        // use mix of C++ and C - API, because the C++ API has no 
        // support for variable length arrays :(
        
        NcFile *file = new NcFile( path.c_str(), NcFile::Replace );
        
        if ( ! file->is_valid() )
        {
            std::cerr << "Error opening output file at " << path << std::endl;
            
            exit(-1);
        }
        
        // Create dimensions
        
        NcDim *dim = file->add_dim("dim", (int) 
featureSpace->coordinate_system->size() );
        NcDim *N = file->add_dim( "N", clusters.size() );
        
        // Create attributes
        
        file->add_att( "num_clusters", (int) clusters.size() );
        file->add_att( "source", featureSpace->filename().c_str() );
        file->add_att( "parameters", parameters.c_str() );
        file->close();

        
        // Open the file with the C API as well

        int retval, ncid;
        
        if ( ! nc_open( path.c_str(), NC_WRITE, &ncid ) )
        {
            cerr << "Error opening file " << path << "for writing" << endl;
            
            exit( -1 );
        }
        
        // Add cluster variables
        
        for ( size_t ci = 0; ci < clusters.size(); ci++ )
        {
            stringstream ss(stringstream::in | stringstream::out);
            
            // Create variable
            ss << "cluster_" << ci;
            
            // Declare a variable length array type
            
            nc_type type_id, base_type;
            size_t type_size;
            char type_name[255];

            nc_def_vlen( ncid, "cluster_type", NC_DOUBLE, &type_id );
            
            nc_inq_vlen( ncid, type_id, &type_name[0], &type_size, &base_type );
            
            // Create a variable with the new VLEN type

            int var_id, dim_id;
            
            nc_inq_dimid( ncid, "dim", &dim_id );
            
            int dims[1] = {dim_id};
            
            nc_def_var( ncid, ss.str().c_str(), type_id, 1, dims, &var_id );
            
            // write attributes mode, size, id
            
            vector<T> mode = clusters[ci].mode;
            
            ss.clear();
            
            // mode
            
            for (size_t i = 0; i < mode.size(); i++ ) 
            { 
                ss << mode[i] << " "; 
            };
            
            const char *mode_values[1] = {ss.str().c_str()};
            
            nc_put_att( ncid, var_id, "mode", NC_STRING, 1, mode_values );
            
            // size
            
            int size_values[1] = {(int) clusters[ci].points.size()}; 
            
            nc_put_att_int( ncid, var_id, "size", NC_INT, 1, size_values );
            
            // id
            
            const char *id_values[1] = {clusters[ci].id.c_str()};
            
            nc_put_att( ncid, var_id, "id", NC_STRING, 1, id_values );
            
            // allocate memory and write the cluster away
            
            double 
data[clusters[ci].points.size()][featureSpace->coordinate_system->size()];

            for ( size_t pi = 0; pi < clusters[ci].points.size(); pi++ )
            {
                Point<T> *p = clusters[ci].points[pi];
                
                for ( size_t di = 0; di < 
featureSpace->coordinate_system->size(); di++ )
                {
                    data[pi][di] = p->coordinate[di];
                }
            }
            
            size_t start=0, len=clusters[ci].points.size();
            
            nc_put_vara_double( ncid, var_id, &start, &len, &data[0][0] );
            
//            var->put( &data[0][0] );
        }
        
        nc_close( ncid );

Thanks in advance!

J.

• Follow-Ups:
  • Re: [netcdfgroup] How to represent this data in NetCDF
    • From: Jürgen Lorenz Simon

• References:
  • [netcdfgroup] How to represent this data in NetCDF
    • From: Jürgen Lorenz Simon
  • Re: [netcdfgroup] How to represent this data in NetCDF
    • From: Dave Allured