issue_comments

we have recently switched to using DataArray as the core data structure underlying the TimeSeries class of Darts (unit8co/darts)

We'd be happy to add this to https://xarray.pydata.org/en/stable/ecosystem.html if you can send in a PR!

Done! https://github.com/pydata/xarray/pull/5819 Thanks.

we have recently switched to using DataArray as the core data structure underlying the TimeSeries class of Darts (unit8co/darts)

We'd be happy to add this to https://xarray.pydata.org/en/stable/ecosystem.html if you can send in a PR!

Yes this is a good point. In fact I'm not really willing to change the index in place, but I'm doing so because I've observed that the freq attribute of the index is dropped after sorting the array (see: https://github.com/pydata/xarray/issues/5466), so I use this hack to restore it.

Btw, we have recently switched to using DataArray as the core data structure underlying the TimeSeries class of Darts (https://github.com/unit8co/darts/), and it's working great for us. Thanks a lot for this super nice lib :)

In your example it is safer to set freq when creating time_index.

Hmm time_index.freq seem to be ignored when passing the index to the DataArray constructor. We should probably fix that.

Beware when you set index attributes in place like xa_copy.get_index('time').freq = freq. Currently, Xarray internals are such that depending on how you access the index it may return distinct (although intially identical) objects, e.g.,

```python

xa_copy.get_index('time').freq = freq xa_copy._indexes['time'].index # freq set here DatetimeIndex(['2013-01-01', '2013-02-01', '2013-03-01', '2013-04-01', '2013-05-01', '2013-06-01', '2013-07-01', '2013-08-01', '2013-09-01', '2013-10-01'], dtype='datetime64[ns]', name='time', freq='MS') xa_copy._coords['time']._data.array # freq not set here DatetimeIndex(['2013-01-01', '2013-02-01', '2013-03-01', '2013-04-01', '2013-05-01', '2013-06-01', '2013-07-01', '2013-08-01', '2013-09-01', '2013-10-01'], dtype='datetime64[ns]', name='time', freq=None) xa_copy._coords['time']._data.array is xa_copy._indexes['time'].index False ```

Hopefully, this will become less likely after the ongoing explicit index refactor in Xarray. But in general I would not advise changing in-place pandas indexes that are already bound to xarray objects. In your example it is safer to set freq when creating time_index.

This is the behavior that we get when directly indexing the pandas.DatetimeIndex:

```python

idx = xa_orig.get_index('time') idx DatetimeIndex(['2013-01-01', '2013-02-01', '2013-03-01', '2013-04-01', '2013-05-01', '2013-06-01', '2013-07-01', '2013-08-01', '2013-09-01', '2013-10-01'], dtype='datetime64[ns]', name='time', freq='MS') idx[0:2] # slice of ints -> preserve freq DatetimeIndex(['2013-01-01', '2013-02-01'], dtype='datetime64[ns]', name='time', freq='MS') idx[[0, 1]] # list of ints -> drop freq DatetimeIndex(['2013-01-01', '2013-02-01'], dtype='datetime64[ns]', name='time', freq=None) ```

I think that this behavior makes sense since indexing the index with an arbitrary list of integer positions may not yield labels with a meaningful, unique frequency. With a slice it can be updated easily (pandas does it), e.g.,

```python

idx[0:2:4] DatetimeIndex(['2013-01-01'], dtype='datetime64[ns]', name='time', freq='4MS') ```

Not sure what's happening with the copy of the DataArray, though.