Requirements for function passed to to DataArrayGroupBy.map()

[jlumpe]

The user documentation doesn't really give much information on what kind of functions you can use with map(), it seems like it only covers numpy ufuncs or compositions of them. The example of using a custom function is:

def standardize(x):
     return (x - x.mean()) / x.std()

which will return a DataArray for DataArray input. The documentation doesn't explicitly state that this is required, but when I try to use it with a function that returns a scalar I get:

~/opt/anaconda3/envs/gambit/lib/python3.9/site-packages/xarray/core/groupby.py in map(self, func, shortcut, args, **kwargs)
    792         grouped = self._iter_grouped_shortcut() if shortcut else self._iter_grouped()
    793         applied = (maybe_wrap_array(arr, func(arr, *args, **kwargs)) for arr in grouped)
--> 794         return self._combine(applied, shortcut=shortcut)
    795 
    796     def apply(self, func, shortcut=False, args=(), **kwargs):

~/opt/anaconda3/envs/gambit/lib/python3.9/site-packages/xarray/core/groupby.py in _combine(self, applied, shortcut)
    812         """Recombine the applied objects like the original."""
    813         applied_example, applied = peek_at(applied)
--> 814         coord, dim, positions = self._infer_concat_args(applied_example)
    815         if shortcut:
    816             combined = self._concat_shortcut(applied, dim, positions)

~/opt/anaconda3/envs/gambit/lib/python3.9/site-packages/xarray/core/groupby.py in _infer_concat_args(self, applied_example)
    460 
    461     def _infer_concat_args(self, applied_example):
--> 462         if self._group_dim in applied_example.dims:
    463             coord = self._group
    464             positions = self._group_indices

AttributeError: 'int' object has no attribute 'dims'

Wrapping the return value in a zero-dimensional DataArray fixes the issue. Is this the expected behavior?

[mathause]

Do you remember the function you used? E.g. the following works

air = xr.tutorial.load_dataset("air_temperature")
air.map(lambda x: 1)

[keewis]

I think the call is something like air.groupby("time.month").map(lambda x: 1) which does indeed raise the posted error. This makes sense to me because in addition to applying the function to the groups it tries to concatenate the result.

[jlumpe]

@keewis is correct, this is using the result of DataArray.groupby(). My expectation is that any return value that is not an instance of DataArray should be converted to one (probably just by calling DataArray(value)), similar to how Pandas would convert the return value of the function passed to DataFrame.apply().