Timedelta dt accessor does not work

[Tillsten]

MCVE Code Sample

# Your code here
import xarray as xr
import datetime as dt

now = dt.datetime.now()
dt_array = xr.DataArray([now + dt.timedelta(seconds=i) for i in range(10)])

dt_array.dt.second #works

start_t = dt_array - dt_array[0]
start_t.dt.second #fails

Expected Output

Last line should give me the time timedelta in seconds. Instead it raises an exception. Same for hour and minute.

Output of xr.show_versions()

INSTALLED VERSIONS ------------------ commit: None python: 3.7.3 (default, Apr 24 2019, 15:29:51) [MSC v.1915 64 bit (AMD64)] python-bits: 64 OS: Windows OS-release: 10 machine: AMD64 processor: Intel64 Family 6 Model 158 Stepping 10, GenuineIntel byteorder: little LC_ALL: None LANG: None LOCALE: None.None libhdf5: None libnetcdf: None

xarray: 0.14.0
pandas: 0.25.3
numpy: 1.17.4
scipy: 1.2.1
netCDF4: None
pydap: None
h5netcdf: None
h5py: None
Nio: None
zarr: None
cftime: None
nc_time_axis: None
PseudoNetCDF: None
rasterio: None
cfgrib: None
iris: None
bottleneck: None
dask: None
distributed: None
matplotlib: 3.1.1
cartopy: None
seaborn: None
numbagg: None
setuptools: 42.0.2.post20191203
pip: 19.1.1
conda: None
pytest: 5.0.0
IPython: 7.7.0
sphinx: 2.2.1

[andersy005]

1. Per Pandas documentation, the following are the defined attributes for Timedelta:

asm8	Return a numpy timedelta64 array scalar view.
components	Return a components namedtuple-like.
days	Number of days.
delta	Return the timedelta in nanoseconds (ns), for internal compatibility.
microseconds	Number of microseconds (>= 0 and less than 1 second).
nanoseconds	Return the number of nanoseconds (n), where 0 <= n < 1 microsecond.
resolution	Return a string representing the lowest timedelta resolution.
resolution_string	Return a string representing the lowest timedelta resolution.
seconds	Number of seconds (>= 0 and less than 1 day).

and since xarray coerces the dataarray to a Pandas Series prior accessing the requested attribute, calling start_t.dt.second fails.

2. Since seconds is a defined attribute for pandas.TimeDelta, I expect start_t.dt.seconds to work. However, it fails:

In [33]: start_t.dt.seconds
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-33-e1e0943720f8> in <module>
----> 1 start_t.dt.seconds

AttributeError: 'DatetimeAccessor' object has no attribute 'seconds'

This is likely a bug in xarray.

3. Pandas appears to be establishing differences between DatetimeProperties and TimedeltaProperties:

In [16]: start_t_series.dt
Out[16]: <pandas.core.indexes.accessors.TimedeltaProperties object at 0x7f2afafd45c0>

In [17]: dt_array_series.dt
Out[17]: <pandas.core.indexes.accessors.DatetimeProperties object at 0x7f2bf7ccd320>

Xarray, on the other hand, seems to be treating them the same:

In [18]: dt_array.dt
Out[18]: <xarray.core.accessor_dt.DatetimeAccessor at 0x7f2afb6e8710>

In [19]: start_t.dt
Out[19]: <xarray.core.accessor_dt.DatetimeAccessor at 0x7f2afb6a20f0>

• Should xarray establish differences between DatetimeProperties and TimedeltaProperties??

[dcherian]

This is more unimplemented feature than a bug. .dt is really a DatetimeAccessor.

The Timedelta functionality hasn't been implemented e.g.

xarray/xarray/core/accessor_dt.py

Line 204 in 2ee89c3

second = _tslib_field_accessor("second", "The seconds of the datetime", np.int64)

[andersy005]

@dcherian, Thank you for chiming in! Are there any other issues with discussions about implementation for Timedelta functionality? I am happy to take a stab at implementing this.

[dcherian]

I think this would be a welcome feature. It seems like there should be a lot of shared code between the two accessors.