Add groupby.dims & Fix groupby reduce for DataArray

doc/groupby.rst

@@ -215,4 +215,4 @@ applying your function, and then unstacking the result:
 .. ipython:: python
 
    stacked = da.stack(gridcell=['ny', 'nx'])
-   stacked.groupby('gridcell').sum().unstack('gridcell')
+   stacked.groupby('gridcell').sum(xr.ALL_DIMS).unstack('gridcell')

doc/whats-new.rst

@@ -24,13 +24,17 @@ New functions/methods
 Enhancements
 ~~~~~~~~~~~~
 
-- Add a repr for :py:class:`~xarray.core.GroupBy` objects. By `Deepak Cherian <https://github.com/dcherian>`_.
-  Example::
+- :py:class:`~xarray.core.GroupBy` enhancements. By `Deepak Cherian <https://github.com/dcherian>`_.
+
+  - Added a repr. Example::
 
       >>> da.groupby("time.season")
       DataArrayGroupBy, grouped over 'season'
       4 groups with labels 'DJF', 'JJA', 'MAM', 'SON'
 
+  - Added a ``GroupBy.dims`` property that mirrors the dimensions
+    of each group.
+
 Bug fixes
 ~~~~~~~~~
 - Reintroduce support for :mod:`weakref` (broken in v0.13.0). Support has been
@@ -40,6 +44,9 @@ Bug fixes
 - Line plots with the ``x`` or ``y`` argument set to a 1D non-dimensional coord
   now plot the correct data for 2D DataArrays.
   (:issue:`3334`). By `Tom Nicholas <http://github.com/TomNicholas>`_.
+- Fix deprecation of default reduction dimension for :py:class:`~xarray.core.groupby.DataArrayGroupBy` objects.
+  (:issue:`3337`). Also make sure that grouping dimensions with ``dtype='object'`` indexes
+  is squeezed just as for other indexes. By `Deepak Cherian <https://github.com/dcherian>`_.
 
 Documentation
 ~~~~~~~~~~~~~

xarray/core/groupby.py

@@ -7,7 +7,7 @@
 
 from . import dtypes, duck_array_ops, nputils, ops
 from .arithmetic import SupportsArithmetic
-from .common import ImplementsArrayReduce, ImplementsDatasetReduce
+from .common import ALL_DIMS, ImplementsArrayReduce, ImplementsDatasetReduce
 from .concat import concat
 from .formatting import format_array_flat
 from .options import _get_keep_attrs
@@ -191,7 +191,10 @@ def _unique_and_monotonic(group):
         return True
     else:
         index = safe_cast_to_index(group)
-        return index.is_unique and index.is_monotonic
+        if index.dtype == "O":
+            return index.is_unique
+        else:
+            return index.is_unique and index.is_monotonic
 
 
 def _apply_loffset(grouper, result):
@@ -248,6 +251,7 @@ class GroupBy(SupportsArithmetic):
         "_restore_coord_dims",
         "_stacked_dim",
         "_unique_coord",
+        "dims",
     )
 
     def __init__(
@@ -380,6 +384,9 @@ def __init__(
         # cached attributes
         self._groups = None
 
+        example = obj.isel(**{group_dim: group_indices[0]})
+        self.dims = example.dims
+
     @property
     def groups(self):
         # provided to mimic pandas.groupby
@@ -689,7 +696,7 @@ def quantile(self, q, dim=None, interpolation="linear", keep_attrs=None):
         q : float in range of [0,1] (or sequence of floats)
             Quantile to compute, which must be between 0 and 1
             inclusive.
-        dim : str or sequence of str, optional
+        dim : xarray.ALL_DIMS, str or sequence of str, optional
             Dimension(s) over which to apply quantile.
             Defaults to the grouped dimension.
         interpolation : {'linear', 'lower', 'higher', 'midpoint', 'nearest'}
@@ -746,7 +753,7 @@ def reduce(
             Function which can be called in the form
             `func(x, axis=axis, **kwargs)` to return the result of collapsing
             an np.ndarray over an integer valued axis.
-        dim : str or sequence of str, optional
+        dim : xarray.ALL_DIMS, str or sequence of str, optional
             Dimension(s) over which to apply `func`.
         axis : int or sequence of int, optional
             Axis(es) over which to apply `func`. Only one of the 'dimension'
@@ -765,9 +772,18 @@ def reduce(
             Array with summarized data and the indicated dimension(s)
             removed.
         """
+        if dim is None:
+            dim = self._group_dim
+
         if keep_attrs is None:
             keep_attrs = _get_keep_attrs(default=False)
 
+        if dim is not ALL_DIMS and dim not in self.dims:
+            raise ValueError(
+                "cannot reduce over dimension %r. expected either xarray.ALL_DIMS to reduce over all dimensions or one or more of %r."
+                % (dim, self.dims)
+            )
+
         def reduce_array(ar):
             return ar.reduce(func, dim, axis, keep_attrs=keep_attrs, **kwargs)
 
@@ -835,7 +851,7 @@ def reduce(self, func, dim=None, keep_attrs=None, **kwargs):
             Function which can be called in the form
             `func(x, axis=axis, **kwargs)` to return the result of collapsing
             an np.ndarray over an integer valued axis.
-        dim : str or sequence of str, optional
+        dim : xarray.ALL_DIMS, str or sequence of str, optional
             Dimension(s) over which to apply `func`.
         axis : int or sequence of int, optional
             Axis(es) over which to apply `func`. Only one of the 'dimension'
@@ -863,6 +879,12 @@ def reduce(self, func, dim=None, keep_attrs=None, **kwargs):
         def reduce_dataset(ds):
             return ds.reduce(func, dim, keep_attrs, **kwargs)
 
+        if dim is not ALL_DIMS and dim not in self.dims:
+            raise ValueError(
+                "cannot reduce over dimension %r. expected either xarray.ALL_DIMS to reduce over all dimensions or one or more of %r."
+                % (dim, self.dims)
+            )
+
         return self.apply(reduce_dataset)
 
     def assign(self, **kwargs):

xarray/tests/test_dataarray.py

@@ -2582,6 +2582,15 @@ def change_metadata(x):
         expected = change_metadata(expected)
         assert_equal(expected, actual)
 
+    def test_groupby_reduce_dimension_error(self):
+        array = self.make_groupby_example_array()
+        grouped = array.groupby("y")
+        with raises_regex(ValueError, "cannot reduce over dimension 'y'"):
+            grouped.mean()
+
+        grouped = array.groupby("y", squeeze=False)
+        assert_identical(array, grouped.mean())
+
     def test_groupby_math(self):
         array = self.make_groupby_example_array()
         for squeeze in [True, False]:

xarray/tests/test_groupby.py

@@ -21,6 +21,22 @@ def test_consolidate_slices():
         _consolidate_slices([slice(3), 4])
 
 
+def test_groupby_dims_property():
+    ds = xr.Dataset(
+        {"foo": (("x", "y", "z"), np.random.randn(3, 4, 2))},
+        {"x": ["a", "b", "c"], "y": [1, 2, 3, 4], "z": [1, 2]},
+    )
+
+    assert ds.groupby("x").dims == ds.sel(x="a").dims
+    assert ds.groupby("y").dims == ds.sel(y=1).dims
+
+    assert ds.groupby("x", squeeze=False).dims == ds.isel(x=[1]).dims
+    assert ds.groupby("y", squeeze=False).dims == ds.isel(y=[1]).dims
+
+    stacked = ds.stack({"xy": ("x", "y")})
+    assert stacked.groupby("xy").dims == stacked.isel(xy=0).dims
+
+
 def test_multi_index_groupby_apply():
     # regression test for GH873
     ds = xr.Dataset(