About the distinction between DataArray.nbytes and Dataset.nbytes

[etienneschalk]

About the distinction between DataArray.nbytes and Dataset.nbytes

I have a question about the current nbytes that is represented, inherited from a distinction between DataArray.nbytes and Dataset.nbytes:

• For a DataArray, only the nbytes of the data itself is taken into account (without coordinates)
• Whereas for a Dataset, all data variables and coordinates variables contribute to the size.

Examples: First I build a DataArray where the coordinates (float64) weigh more than the actual data (uint8). The top-level Size: 3B is shown. Then I place it into a Dataset. Now Size: 27B is shown. This different way of counting is inherited from the direct usage of .nbytes. I find it a bit confusing because the DataArray and Dataset contain the same data, but presented differently.

The documentation is clear about it:

• DataArray.nbytes Total bytes consumed by the elements of this DataArray’s data.
• Dataset.nbytes Total bytes consumed by the data arrays of all variables in this dataset.

But still, is there any reason for this variation? Why excluding coordinates for DataArrays and not for Datasets?

Thanks!

code examples below ⬇️

In [14]: xda = xr.DataArray(data=np.array([11,22,33], dtype=np.uint8), coords=dict(x=np.array([10,20,30], dtype=np.float64)))

In [15]: xda
Out[15]: 
<xarray.DataArray (x: 3)> Size: 3B
array([11, 22, 33], dtype=uint8)
Coordinates:
  * x        (x) float64 24B 10.0 20.0 30.0

In [19]: xds = xr.Dataset({'myvar': xda})

In [20]: xds
Out[20]: 
<xarray.Dataset> Size: 27B
Dimensions:  (x: 3)
Coordinates:
  * x        (x) float64 24B 10.0 20.0 30.0
Data variables:
    myvar    (x) uint8 3B 11 22 33

In [21]: xda.nbytes
Out[21]: 3

In [22]: xds.nbytes
Out[22]: 27