Merge branch 'master' into ml-evs/fix_#37098

doc/source/whatsnew/v1.2.0.rst

@@ -252,8 +252,115 @@ Other enhancements
 - :func:`read_csv` supports memory-mapping for compressed files (:issue:`37621`)
 - Improve error reporting for :meth:`DataFrame.merge()` when invalid merge column definitions were given (:issue:`16228`)
 - Improve numerical stability for :meth:`Rolling.skew()`, :meth:`Rolling.kurt()`, :meth:`Expanding.skew()` and :meth:`Expanding.kurt()` through implementation of Kahan summation (:issue:`6929`)
+- Improved error reporting for subsetting columns of a :class:`DataFrameGroupBy` with ``axis=1`` (:issue:`37725`)
 - :func:`pandas.load_pickle` can now load from ``.zip`` files created by OS X/macOS that contain ``__MACOSX/`` or ``.DS_STORE`` hidden folders/files (:issue:`37098`).
 
+
+.. ---------------------------------------------------------------------------
+
+.. _whatsnew_120.notable_bug_fixes:
+
+Notable bug fixes
+~~~~~~~~~~~~~~~~~
+
+These are bug fixes that might have notable behavior changes.
+
+Consistency of DataFrame Reductions
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+:meth:`DataFrame.any` and :meth:`DataFrame.all` with ``bool_only=True`` now
+determines whether to exclude object-dtype columns on a column-by-column basis,
+instead of checking if *all* object-dtype columns can be considered boolean.
+
+This prevents pathological behavior where applying the reduction on a subset
+of columns could result in a larger :class:`Series` result. See (:issue:`37799`).
+
+.. ipython:: python
+
+    df = pd.DataFrame({"A": ["foo", "bar"], "B": [True, False]}, dtype=object)
+    df["C"] = pd.Series([True, True])
+
+
+*Previous behavior*:
+
+.. code-block:: ipython
+
+    In [5]: df.all(bool_only=True)
+    Out[5]:
+    C    True
+    dtype: bool
+
+    In [6]: df[["B", "C"]].all(bool_only=True)
+    Out[6]:
+    B    False
+    C    True
+    dtype: bool
+
+*New behavior*:
+
+.. ipython:: python
+
+    In [5]: df.all(bool_only=True)
+
+    In [6]: df[["B", "C"]].all(bool_only=True)
+
+
+Other :class:`DataFrame` reductions with ``numeric_only=None`` will also avoid
+this pathological behavior (:issue:`37827`):
+
+.. ipython:: python
+
+    df = pd.DataFrame({"A": [0, 1, 2], "B": ["a", "b", "c"]}, dtype=object)
+
+
+*Previous behavior*:
+
+.. code-block:: ipython
+
+    In [3]: df.mean()
+    Out[3]: Series([], dtype: float64)
+
+    In [4]: df[["A"]].mean()
+    Out[4]:
+    A    1.0
+    dtype: float64
+
+*New behavior*:
+
+.. ipython:: python
+
+    df.mean()
+
+    df[["A"]].mean()
+
+Moreover, :class:`DataFrame` reductions with ``numeric_only=None`` will now be
+consistent with their :class:`Series` counterparts.  In particular, for
+reductions where the :class:`Series` method raises ``TypeError``, the
+:class:`DataFrame` reduction will now consider that column non-numeric
+instead of casting to NumPy which may have different semantics (:issue:`36076`,
+:issue:`28949`, :issue:`21020`).
+
+.. ipython:: python
+
+    ser = pd.Series([0, 1], dtype="category", name="A")
+    df = ser.to_frame()
+
+
+*Previous behavior*:
+
+.. code-block:: ipython
+
+    In [5]: df.any()
+    Out[5]:
+    A    True
+    dtype: bool
+
+*New behavior*:
+
+.. ipython:: python
+
+    df.any()
+
+
 .. _whatsnew_120.api_breaking.python:
 
 Increased minimum version for Python