All dimmensions become coordinates?

[wilhelmsen]

I have a netcdf file that looks like this (ncdump):

netcdf filevYpcG2 {
dimensions:
    coor1 = 2 ;
    dim1 = 2 ;
variables:
    int coor1(coor1) ;
        coor1:_FillValue = 999999 ;
    int var1(coor1) ;
        var1:_FillValue = 999999 ;
    int var2(dim1) ;
        var2:units = "unit1" ;
        var2:_FillValue = 999999 ;
        var2:attr_name = "variable1" ;

// global attributes:
        :metadata1 = "metadata1" ;
data:

 coor1 = 1, 2 ;

 var1 = 1, 2 ;

 var2 = 3, 4 ;
}

When opening it in xray (xarray) I therefore expect dim1 to be a dimension and coor1 to be a coordinate, because coor1 is also a variable, coor1(coor1)? But when opening the dataset in xray, I get the following:

In [1]: import xray;

In [2]: filename = "/tmp/filevYpcG2"

In [3]: x = xray.open_dataset(filename)

In [4]: x
Out[4]: 
<xray.Dataset>
Dimensions:  (coor1: 2, dim1: 2)
Coordinates:
  * coor1    (coor1) float64 1.0 2.0
  * dim1     (dim1) int64 0 1
Data variables:
    var1     (coor1) float64 1.0 2.0
    var2     (dim1) float64 3.0 4.0
Attributes:
    metadata1: metadata1


In [5]: x.coords
Out[5]: 
Coordinates:
  * coor1    (coor1) float64 1.0 2.0
  * dim1     (dim1) int64 0 1

In [6]: x.dims
Out[6]: Frozen(SortedKeysDict({u'coor1': 2, u'dim1': 2}))

In [7]: x.data_vars
Out[7]: 
Data variables:
    var1     (coor1) float64 1.0 2.0
    var2     (dim1) float64 3.0 4.0

That is, both coor1 and dim1 becomes coordinate variables. I expected only coor1 to become a coordinate variable, and not dim1.

Am I wrohg, or is it meant to be like this?

[shoyer]

This is definitely intended behavior: if you don't supply a coordinate to label points along a dimension, a default coordinate (equivalently to range(n)) is created for you. This is somewhat similar to pandas. Looking through the docs, though, it looks like we never state this explicitly...

[wilhelmsen]

Thank you!

I was just wondering how you know then, which one is an "actual" coordinate and which one is an indexed ("default") coordinate?

xarray seem to know this.

Example:

In [1]: import xray as xr

In [2]: import numpy as np

In [3]: var1 = xr.DataArray(np.array([1, 2], np.float64), coords=[np.array([1, 2], np.int64)], dims=["coor1",], name = "var1")

In [4]: var2 = xr.DataArray(np.array([3, 4], np.float64), dims=["dim1",], name = "var2")

In [5]: ds.var1
Out[5]: 
<xray.DataArray 'var1' (coor1: 2)>
array([ 1.,  2.])
Coordinates:
  * coor1    (coor1) int64 1 2

In [6]: ds.var2
Out[6]: 
<xray.DataArray 'var2' (dim1: 2)>
array([ 3.,  4.])
Coordinates:
  * dim1     (dim1) int64 0 1

In [7]: ds.var1.coords
Out[7]: 
Coordinates:
  * coor1    (coor1) int64 1 2

In [8]: ds.var2.coords
Out[8]: 
Coordinates:
  * dim1     (dim1) int64 0 1

In [9]: ds.to_netcdf("/tmp/from_xr.nc")

And then

$ ncdump /tmp/from_xr.nc 
netcdf from_xr {
dimensions:
    coor1 = 2 ;
    dim1 = 2 ;
variables:
    int64 coor1(coor1) ;
    double var1(coor1) ;
    double var2(dim1) ;
data:

 coor1 = 1, 2 ;

 var1 = 1, 2 ;

 var2 = 3, 4 ;
}

As dim1 is not written in the file, xarray somehow knows?

[shoyer]

Yes, xarray checks for "trivial indexes" that are equivalent to what it would create automatically when it writes a file to disk:
https://github.com/pydata/xarray/blob/v0.7.0/xarray/backends/common.py#L31-L47

[jhamman]

I'm closing since #1017 has moved us past this question.