Backport PR #43854: REGR: RollingGroupby.corr producing incorrect results (#43859)

Co-authored-by: Richard Shadrach <[email protected]>

Author: meeseeksmachine

doc/source/whatsnew/v1.3.4.rst

@@ -24,6 +24,8 @@ Fixed regressions
 - Fixed regression in :meth:`pandas.read_csv` raising ``UnicodeDecodeError`` exception when ``memory_map=True`` (:issue:`43540`)
 - Fixed regression in :meth:`DataFrame.explode` raising ``AssertionError`` when ``column`` is any scalar which is not a string (:issue:`43314`)
 - Fixed regression in :meth:`Series.aggregate` attempting to pass ``args`` and ``kwargs`` multiple times to the user supplied ``func`` in certain cases (:issue:`43357`)
+- Fixed regression when iterating over a :class:`DataFrame.groupby.rolling` object causing the resulting DataFrames to have an incorrect index if the input groupings were not sorted (:issue:`43386`)
+- Fixed regression in :meth:`DataFrame.groupby.rolling.cov` and :meth:`DataFrame.groupby.rolling.corr` computing incorrect results if the input groupings were not sorted (:issue:`43386`)
 
 .. ---------------------------------------------------------------------------
 

pandas/core/window/rolling.py

@@ -290,8 +290,8 @@ def __repr__(self) -> str:
         return f"{type(self).__name__} [{attrs}]"
 
     def __iter__(self):
-        obj = self._create_data(self._selected_obj)
-        obj = obj.set_axis(self._on)
+        obj = self._selected_obj.set_axis(self._on)
+        obj = self._create_data(obj)
         indexer = self._get_window_indexer()
 
         start, end = indexer.get_window_bounds(
@@ -649,6 +649,7 @@ def _apply_pairwise(
         """
         # Manually drop the grouping column first
         target = target.drop(columns=self._grouper.names, errors="ignore")
+        target = self._create_data(target)
         result = super()._apply_pairwise(target, other, pairwise, func)
         # 1) Determine the levels + codes of the groupby levels
         if other is not None:

pandas/tests/window/test_groupby.py

@@ -146,6 +146,42 @@ def func(x):
         expected = g.apply(func)
         tm.assert_series_equal(result, expected)
 
+    @pytest.mark.parametrize(
+        "func, expected_values",
+        [("cov", [[1.0, 1.0], [1.0, 4.0]]), ("corr", [[1.0, 0.5], [0.5, 1.0]])],
+    )
+    def test_rolling_corr_cov_unordered(self, func, expected_values):
+        # GH 43386
+        df = DataFrame(
+            {
+                "a": ["g1", "g2", "g1", "g1"],
+                "b": [0, 0, 1, 2],
+                "c": [2, 0, 6, 4],
+            }
+        )
+        rol = df.groupby("a").rolling(3)
+        result = getattr(rol, func)()
+        expected = DataFrame(
+            {
+                "b": 4 * [np.nan] + expected_values[0] + 2 * [np.nan],
+                "c": 4 * [np.nan] + expected_values[1] + 2 * [np.nan],
+            },
+            index=MultiIndex.from_tuples(
+                [
+                    ("g1", 0, "b"),
+                    ("g1", 0, "c"),
+                    ("g1", 2, "b"),
+                    ("g1", 2, "c"),
+                    ("g1", 3, "b"),
+                    ("g1", 3, "c"),
+                    ("g2", 1, "b"),
+                    ("g2", 1, "c"),
+                ],
+                names=["a", None, None],
+            ),
+        )
+        tm.assert_frame_equal(result, expected)
+
     def test_rolling_apply(self, raw):
         g = self.frame.groupby("A")
         r = g.rolling(window=4)

pandas/tests/window/test_rolling.py

@@ -792,6 +792,15 @@ def test_iter_rolling_on_dataframe(expected, window):
         tm.assert_frame_equal(actual, expected)
 
 
+def test_iter_rolling_on_dataframe_unordered():
+    # GH 43386
+    df = DataFrame({"a": ["x", "y", "x"], "b": [0, 1, 2]})
+    results = list(df.groupby("a").rolling(2))
+    expecteds = [df.iloc[idx, [1]] for idx in [[0], [0, 2], [1]]]
+    for result, expected in zip(results, expecteds):
+        tm.assert_frame_equal(result, expected)
+
+
 @pytest.mark.parametrize(
     "ser,expected,window, min_periods",
     [