xarray.Dataset.from_array

[raybellwaves]

I would be interested in a xarray.Dataset.from_numpy method.

This would act as a wrapper for xr.DataArray(arr).to_dataset("arr") or a better default variable name than "arr".

Motivation here is a one-line inverse transform of ds.to_numpy().

Obviously this has draw backs compared to building up the xarray object using xr.DataArray and using its args. However, I feel would speed up the process of a using Dataset operations on a numpy array.

[TomNicholas]

Hi @raybellwaves, thanks for the suggestion.

Motivation here is a one-line inverse transform of ds.to_numpy().

I'm a bit confused though - this is already a one-line conversion? You could of course make it neater by locally defining

def numpy_array_to_ds(arr, name="arr"):
    return xr.DataArray(arr).to_dataset(name=name)

I suppose we could perhaps try to further generalise Dataset.__init__, although it is already quite a complicated constructor...

Motivation here is a one-line inverse transform of ds.to_numpy().

(FYI the original motivation for ds.to_numpy() was situations where you need to strip the array of any other duck-type wrappers, for instance pint units. I feel like that isn't really related to what you're asked for here though - what you're asking about is closer to an inverse of .values().)

[raybellwaves]

Thanks for explaining the motivation/purpose of .to_numpy() @TomNicholas. Yeah I probably want xarray.Dataset.from_array as the inverse of ds[key].values. Motivation of the question was when I was going through xarray data format transformations e.g. to_dataframe and from_dataframe. There probably a balance between creating new methods which are simplified chained methods of the current API. Not high priority.

[raybellwaves]

Similar to #5795 (reply in thread) I agree it's best not duplicate existing methods for future maintenance.