issue_comments

Thanks @dcherian!

I just wanted to illustrate that these are the same problems and that only removing the property from the Dataset may lead to continued confusion.

I think that's a good point. There are a few methods on DataArrays but not Datasets, but not many:

python In [4]: {x for x in set(dir(xr.DataArray)) - set(dir(xr.Dataset)) if not x.startswith("_")} Out[4]: {'data', 'dot', 'dt', 'dtype', 'from_cdms2', 'from_iris', 'from_series', 'get_axis_num', 'item', 'name', 'ndim', 'plot', 'searchsorted', 'shape', 'size', 'to_cdms2', 'to_dataset', 'to_index', 'to_iris', 'to_masked_array', 'to_pandas', 'to_series', 'variable'}

Regardless, sounds like @shoyer thinks we should merge (and I'm probably still +0.1)

I suppose the case for removing DataArray.T is that you don't really need a shortcut for transposing arrays in xarray, since most functions don't rely on dimension order. And of course the real motivation: some weather/climate scientists like to use a variable named T for time :).

The counter-argument would be that T is found on most other numpy array-like types, including pandas.DataFrame. I've definitely used .T as a convenient shortcut myself in some cases.

I guess that puts me at -0 for deprecating DataArray.T. But we can merge this already, and leave that for a separate discussion.

Sure, that's a good point. I just wanted to illustrate that these are the same problems and that only removing the property from the Dataset may lead to continued confusion. @shoyer thoughts?

I'm +0.2 for merging this, but at least issuing a DeprecationWarning before removing from DataArray.T (with no strong view on whether we do remove it)

@jhamman one hiccup is that we haven't been issuing a DeprecationWarning for DataArray.T

I'm OK with removing it though.

Feeling like playing devils advocate here. I would expect we'll run into similar confusion with the DataArray transpose property.

```Python

da = xr.DataArray([1, 2, 3, 4], dims=('time', ), coords={'time': pd.date_range('2018-10-24', periods=4)}) da.time

<xarray.DataArray 'time' (time: 4)> array(['2018-10-24T00:00:00.000000000', '2018-10-25T00:00:00.000000000', '2018-10-26T00:00:00.000000000', '2018-10-27T00:00:00.000000000'], dtype='datetime64[ns]') Coordinates: * time (time) datetime64[ns] 2018-10-24 2018-10-25 2018-10-26 2018-10-27

da = xr.DataArray([1, 2, 3, 4], dims=('T', ), coords={'T': pd.date_range('2018-10-24', periods=4)}) da.T

<xarray.DataArray (T: 4)> array([1, 2, 3, 4]) Coordinates: * T (T) datetime64[ns] 2018-10-24 2018-10-25 2018-10-26 2018-10-27 ```

I think this is arguably still useful/expected since DataArray is modeled on a NumPy array.

I agree

For posterity, what did we decide to do with the DataArray.T property?

I think this is arguably still useful/expected since DataArray is modeled on a NumPy array. But I could probably be convinced otherwise...

For posterity, what did we decide to do with the DataArray.T property?

I'm not sure but I've fixed the whats-new entry to say that only Dataset.T has been removed currently.

For posterity, what did we decide to do with the DataArray.T property?