Resample interpolate failing on tutorial dataset

[jhamman]

I'm getting some unexpected behavior/errors from the new resample/interpolate methods.

@darothen - any idea what's going on here?

In [1]: import xarray as xr

In [2]: ds = xr.tutorial.load_dataset('air_temperature')

In [3]: ds.resample(time='15d').interpolate(kind='linear')
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-3-ef931d7ebbda> in <module>()
----> 1 ds.resample(time='15d').interpolate(kind='linear')

/glade/p/work/jhamman/storylines/src/xarray/xarray/core/resample.py in interpolate(self, kind)
    110
    111         """
--> 112         return self._interpolate(kind=kind)
    113
    114     def _interpolate(self, kind='linear'):

/glade/p/work/jhamman/storylines/src/xarray/xarray/core/resample.py in _interpolate(self, kind)
    312
    313         old_times = self._obj[self._dim].astype(float)
--> 314         new_times = self._full_index.values.astype(float)
    315
    316         data_vars = OrderedDict()

AttributeError: 'NoneType' object has no attribute 'values'

[darothen]

(sorry, originally commented from my work account)

The tutorial dataset is ~6-hourly, so your operation is a downsampling operation. We don't actually support interpolation on downsampling operations - just aggregations/reductions. Upsampling supports interpolation since there is no implicit way to estimate data between the gaps at the lower temporal frequency. If you just want to estimate a given field at 15-day intervals, for 00Z on those days, then I think you should use ds.reindex(), but at the moment I do not think it will work with timeseries. That would be a critical feature to implement.

[jhamman]

@darothen - Thanks and interesting. I'm getting the above error in a real world resample operation so I figured they were the same issue. I'll dig into this and add some more detail in a bit.

[jhamman]

Okay, I got it now. Consider this example

dates = pd.date_range('2016-01-01', '2016-12-31', freq='D')
# orig = dates[dates != '2016-02-29'] # drop feb 29 and this example will work
orig = dates

da = xr.DataArray(np.random.random((len(orig), 2, 3)), dims=('time', 'x', 'y'), coords={'time': orig})
print(da)

da.resample(time='1D').interpolate('linear')

<xarray.DataArray (time: 366, x: 2, y: 3)>
array([[[ 0.390107,  0.257026,  0.155619],
        [ 0.151772,  0.98012 ,  0.61582 ]],

       [[ 0.081488,  0.038706,  0.627044],
        [ 0.840926,  0.778831,  0.102756]],

       ..., 
       [[ 0.94791 ,  0.274371,  0.582416],
        [ 0.544428,  0.351174,  0.603062]],

       [[ 0.166722,  0.507593,  0.841115],
        [ 0.099317,  0.649383,  0.842175]]])
Coordinates:
  * time     (time) datetime64[ns] 2016-01-01 2016-01-02 2016-01-03 ...
Dimensions without coordinates: x, y
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-12-e187a9baeec8> in <module>()
      6 print(da)
      7 
----> 8 da.resample(time='1d').interpolate('linear')

~/Dropbox/src/xarray/xarray/core/resample.py in interpolate(self, kind)
    110 
    111         """
--> 112         return self._interpolate(kind=kind)
    113 
    114     def _interpolate(self, kind='linear'):

~/Dropbox/src/xarray/xarray/core/resample.py in _interpolate(self, kind)
    204         f = interp1d(x, y, kind=kind, axis=axis, bounds_error=True,
    205                      assume_sorted=True)
--> 206         new_x = self._full_index.values.astype('float')
    207 
    208         # construct new up-sampled DataArray

AttributeError: 'NoneType' object has no attribute 'values'

The application here is that I'm doing a QC check on a dataset that is sometimes missing Feb 29. It is sufficient for my application to always resample and fill Feb 29 when its missing. The pandas equivalent works:

s = pd.Series(np.random.random((len(orig))), index=orig)
new = s.resample('1D').interpolate('linear')
new.equals(s)

True

I think I have a fix for this which I'll push up quickly.