Bug (with fix?) in month name stitching for aggregate_by()

[niallrobinson]

There is a bug in a special case of aggregated_by. Here seems to be an example case:
If you aggregate_by season on a cube where there are text month coordinates and the cube only covers one season, then iris fails when it stitches the component month names together as there is only one _slices_by_key and so the iterate fails.

>>> print test_data
air temperature / (K)               (time: 204; projection_y_coordinate: 277; projection_x_coordinate: 349)
     Dimension coordinates:
          time                           x                             -                             -
          projection_y_coordinate        -                             x                             -
          projection_x_coordinate        -                             -                             x
     Auxiliary coordinates:
          month                          x                             -                             -
          season                         x                             -                             -
     Scalar coordinates:
          altitude: 10 m
          forecast_period: 0.0 hours, bound=(0.0, 0.0) hours
          originating_centre: US National Weather Service, National Centres for Environmental Predic...
     Cell methods:
          mean: time

>>>_ = test_data.extract(iris.Constraint(month=['Dec', 'Jan', 'Feb']))
>>>_.aggregated_by('season', iris.analysis.MEAN)

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-47-1f9326d6dafb> in <module>()
      1 _ = test_data.extract(iris.Constraint(month=['Dec', 'Jan', 'Feb']))
----> 2 _.aggregated_by('season', iris.analysis.MEAN)

/project/avd/iris/live/testing/iris/cube.pyc in aggregated_by(self, coords, aggregator, **kwargs)
   2951         key = [slice(None, None)] * self.ndim
   2952         # Generate unique index tuple key to maintain monotonicity.
-> 2953         key[dimension_to_groupby] = tuple(range(len(groupby)))
   2954         key = tuple(key)
   2955         aggregateby_cube = self[key]

/project/avd/iris/live/testing/iris/analysis/__init__.pyc in __len__(self)
   1456             value = len(self._slices_by_key)
   1457         else:
-> 1458             value = len([s for s in self.group()])
   1459 
   1460         return value

/project/avd/iris/live/testing/iris/analysis/__init__.pyc in group(self)
   1334                 self._compute_groupby_coords()
   1335                 # Calculate the new shared coordinates.
-> 1336                 self._compute_shared_coords()
   1337             # Generate the group-by slices/groups.
   1338             for groupby_slice in self._slices_by_key.itervalues():

/project/avd/iris/live/testing/iris/analysis/__init__.pyc in _compute_shared_coords(self)
   1402                     new_bounds = None
   1403                     for key_slice in self._slices_by_key.itervalues():
-> 1404                         new_pt = '|'.join(coord.points[i] for i in key_slice)
   1405                         new_points.append(new_pt)
   1406                 else:

TypeError: 'slice' object is not iterable

> /project/avd/iris/live/testing/iris/analysis/__init__.py(1404)_compute_shared_coords()
   1403                     for key_slice in self._slices_by_key.itervalues():
-> 1404                         new_pt = '|'.join(coord.points[i] for i in key_slice)
   1405                         new_points.append(new_pt)

Which disappears as soon as you do this

>>>_ = test_data.extract(iris.Constraint(month=['Dec', 'Jan', 'Feb', 'Mar']))
>>>_.aggregated_by('season', iris.analysis.MEAN)
<iris 'Cube' of air temperature / (K) (-- : 2; projection_y_coordinate: 277; projection_x_coordinate: 349)>

or this

>>>_ = test_data.extract(iris.Constraint(month=['Dec', 'Jan', 'Feb']))
>>>_.remove_coord('month')
>>>_.aggregated_by('season', iris.analysis.MEAN)
<iris 'Cube' of air temperature / (K) (time: 1; projection_y_coordinate: 277; projection_x_coordinate: 349)>

This might be fixed merely by changing the code to this:

       # Create new shared bounded coordinates.
        for coord in self._shared_coords:
            if coord.points.dtype.kind == 'S':
                if coord.bounds is None:
                    new_points = []
                    new_bounds = None
                    for key_slice in self._slices_by_key.itervalues():
                        try:
                            new_pt = '|'.join(coord.points[i] for i in key_slice)
                        except TypeError:
                            new_pt = '|'.join(coord.points[key_slice])
                        new_points.append(new_pt)
                else:
                    msg = ('collapsing the bounded string coordinate {0!r}'
                           ' is not supported'.format(coord.name()))
                    raise ValueError(msg)
            else:
                new_bounds = []

[pelson]

@niallrobinson - I can't reproduce this. My code is:

import iris
import numpy as np

aux_coords = [(iris.coords.AuxCoord(list('aba'), long_name='foo'), 0)]

cube = iris.cube.Cube(np.arange(12).reshape(3, 4),
                      aux_coords_and_dims=aux_coords)

print cube
print cube.aggregated_by('foo', iris.analysis.MEAN)
print cube[:2].aggregated_by('foo', iris.analysis.MEAN)
print cube[:1].aggregated_by('foo', iris.analysis.MEAN)

Have I missed a key part to this? What version are you using?

[pelson]

Have I missed a key part to this? What version are you using?

Scrap that. I've just added another coordinate ((iris.coords.AuxCoord(list('aaa'), long_name='bar'), 0),) and I got the exception.

[ajdawson]

Fixed by #1208.