COMPAT: Emit warning when groupby by a tuple (#18731)

* COMPAT: Emit warning when groupby by a tuple

Closes #18314

* DOC: avoid future warning

* Cleanup, test unhashable

* PEP8

* Correct KeyError

* update

* xfail

* remove old comments

* pep8

* Fixups

Author: TomAugspurger

doc/source/groupby.rst

@@ -1091,7 +1091,7 @@ You can also select multiple rows from each group by specifying multiple nth val
    business_dates = pd.date_range(start='4/1/2014', end='6/30/2014', freq='B')
    df = pd.DataFrame(1, index=business_dates, columns=['a', 'b'])
    # get the first, 4th, and last date index for each month
-   df.groupby((df.index.year, df.index.month)).nth([0, 3, -1])
+   df.groupby([df.index.year, df.index.month]).nth([0, 3, -1])
 
 Enumerate group items
 ~~~~~~~~~~~~~~~~~~~~~

doc/source/whatsnew/v0.22.0.txt

@@ -203,6 +203,10 @@ Deprecations
 - ``Series.from_array`` and ``SparseSeries.from_array`` are deprecated. Use the normal constructor ``Series(..)`` and ``SparseSeries(..)`` instead (:issue:`18213`).
 - ``DataFrame.as_matrix`` is deprecated. Use ``DataFrame.values`` instead (:issue:`18458`).
 - ``Series.asobject``, ``DatetimeIndex.asobject``, ``PeriodIndex.asobject`` and ``TimeDeltaIndex.asobject`` have been deprecated. Use ``.astype(object)`` instead (:issue:`18572`)
+- Grouping by a tuple of keys now emits a ``FutureWarning`` and is deprecated.
+  In the future, a tuple passed to ``'by'`` will always refer to a single key
+  that is the actual tuple, instead of treating the tuple as multiple keys. To
+  retain the previous behavior, use a list instead of a tuple (:issue:`18314`)
 - ``Series.valid`` is deprecated. Use :meth:`Series.dropna` instead (:issue:`18800`).
 
 .. _whatsnew_0220.prior_deprecations:

pandas/core/groupby.py

@@ -28,6 +28,7 @@
     is_bool_dtype,
     is_scalar,
     is_list_like,
+    is_hashable,
     needs_i8_conversion,
     _ensure_float64,
     _ensure_platform_int,
@@ -2850,7 +2851,27 @@ def _get_grouper(obj, key=None, axis=0, level=None, sort=True,
     elif isinstance(key, BaseGrouper):
         return key, [], obj
 
-    # Everything which is not a list is a key (including tuples):
+    # In the future, a tuple key will always mean an actual key,
+    # not an iterable of keys. In the meantime, we attempt to provide
+    # a warning. We can assume that the user wanted a list of keys when
+    # the key is not in the index. We just have to be careful with
+    # unhashble elements of `key`. Any unhashable elements implies that
+    # they wanted a list of keys.
+    # https://github.com/pandas-dev/pandas/issues/18314
+    is_tuple = isinstance(key, tuple)
+    all_hashable = is_tuple and is_hashable(key)
+
+    if is_tuple:
+        if ((all_hashable and key not in obj and set(key).issubset(obj))
+                or not all_hashable):
+            # column names ('a', 'b') -> ['a', 'b']
+            # arrays like (a, b) -> [a, b]
+            msg = ("Interpreting tuple 'by' as a list of keys, rather than "
+                   "a single key. Use 'by=[...]' instead of 'by=(...)'. In "
+                   "the future, a tuple will always mean a single key.")
+            warnings.warn(msg, FutureWarning, stacklevel=5)
+            key = list(key)
+
     if not isinstance(key, list):
         keys = [key]
         match_axis_length = False

pandas/tests/groupby/test_groupby.py

@@ -2727,6 +2727,38 @@ def test_empty_dataframe_groupby(self):
 
         assert_frame_equal(result, expected)
 
+    def test_tuple_warns(self):
+        # https://github.com/pandas-dev/pandas/issues/18314
+        df = pd.DataFrame({('a', 'b'): [1, 1, 2, 2], 'a': [1, 1, 1, 2],
+                           'b': [1, 2, 2, 2], 'c': [1, 1, 1, 1]})
+        with tm.assert_produces_warning(FutureWarning) as w:
+            df[['a', 'b', 'c']].groupby(('a', 'b')).c.mean()
+
+        assert "Interpreting tuple 'by' as a list" in str(w[0].message)
+
+        with tm.assert_produces_warning(None):
+            df.groupby(('a', 'b')).c.mean()
+
+    def test_tuple_warns_unhashable(self):
+        # https://github.com/pandas-dev/pandas/issues/18314
+        business_dates = date_range(start='4/1/2014', end='6/30/2014',
+                                    freq='B')
+        df = DataFrame(1, index=business_dates, columns=['a', 'b'])
+
+        with tm.assert_produces_warning(FutureWarning) as w:
+            df.groupby((df.index.year, df.index.month)).nth([0, 3, -1])
+
+        assert "Interpreting tuple 'by' as a list" in str(w[0].message)
+
+    @pytest.mark.xfail(reason="GH-18798")
+    def test_tuple_correct_keyerror(self):
+        # https://github.com/pandas-dev/pandas/issues/18798
+        df = pd.DataFrame(1, index=range(3),
+                          columns=pd.MultiIndex.from_product([[1, 2],
+                                                              [3, 4]]))
+        with tm.assert_raises_regex(KeyError, "(7, 8)"):
+            df.groupby((7, 8)).mean()
+
 
 def _check_groupby(df, result, keys, field, f=lambda x: x.sum()):
     tups = lmap(tuple, df[keys].values)