fix Rolling for multi-index and reversed index. (#28297)

Author: leftys

doc/source/whatsnew/v1.0.0.rst

@@ -372,8 +372,9 @@ Plotting
 Groupby/resample/rolling
 ^^^^^^^^^^^^^^^^^^^^^^^^
 
--
 - Bug in :meth:`DataFrame.rolling` not allowing for rolling over datetimes when ``axis=1`` (:issue: `28192`)
+- Bug in :meth:`DataFrame.rolling` not allowing rolling over multi-index levels (:issue: `15584`).
+- Bug in :meth:`DataFrame.rolling` not allowing rolling on monotonic decreasing time indexes (:issue: `19248`).
 - Bug in :meth:`DataFrame.groupby` not offering selection by column name when ``axis=1`` (:issue:`27614`)
 - Bug in :meth:`DataFrameGroupby.agg` not able to use lambda function with named aggregation (:issue:`27519`)
 - Bug in :meth:`DataFrame.groupby` losing column name information when grouping by a categorical column (:issue:`28787`)

pandas/core/window/rolling.py

@@ -70,7 +70,7 @@ def __init__(
         center: Optional[bool] = False,
         win_type: Optional[str] = None,
         axis: Axis = 0,
-        on: Optional[str] = None,
+        on: Optional[Union[str, Index]] = None,
         closed: Optional[str] = None,
         **kwargs
     ):
@@ -126,7 +126,7 @@ def _create_blocks(self):
         obj = self._selected_obj
 
         # filter out the on from the object
-        if self.on is not None:
+        if self.on is not None and not isinstance(self.on, Index):
             if obj.ndim == 2:
                 obj = obj.reindex(columns=obj.columns.difference([self.on]), copy=False)
         blocks = obj._to_dict_of_blocks(copy=False).values()
@@ -637,10 +637,10 @@ class Window(_Window):
         Provide a window type. If ``None``, all points are evenly weighted.
         See the notes below for further information.
     on : str, optional
-        For a DataFrame, a datetime-like column on which to calculate the rolling
-        window, rather than the DataFrame's index. Provided integer column is
-        ignored and excluded from result since an integer index is not used to
-        calculate the rolling window.
+        For a DataFrame, a datetime-like column or MultiIndex level on which
+        to calculate the rolling window, rather than the DataFrame's index.
+        Provided integer column is ignored and excluded from result since
+        an integer index is not used to calculate the rolling window.
     axis : int or str, default 0
     closed : str, default None
         Make the interval closed on the 'right', 'left', 'both' or
@@ -1651,18 +1651,19 @@ def is_datetimelike(self):
 
     @cache_readonly
     def _on(self):
-
         if self.on is None:
             if self.axis == 0:
                 return self.obj.index
             elif self.axis == 1:
                 return self.obj.columns
+        elif isinstance(self.on, Index):
+            return self.on
         elif isinstance(self.obj, ABCDataFrame) and self.on in self.obj.columns:
             return Index(self.obj[self.on])
         else:
             raise ValueError(
                 "invalid on specified as {0}, "
-                "must be a column (if DataFrame) "
+                "must be a column (of DataFrame), an Index "
                 "or None".format(self.on)
             )
 
@@ -1706,10 +1707,12 @@ def validate(self):
 
     def _validate_monotonic(self):
         """
-        Validate on is_monotonic.
+        Validate monotonic (increasing or decreasing).
         """
-        if not self._on.is_monotonic:
-            formatted = self.on or "index"
+        if not (self._on.is_monotonic_increasing or self._on.is_monotonic_decreasing):
+            formatted = self.on
+            if self.on is None:
+                formatted = "index"
             raise ValueError("{0} must be monotonic".format(formatted))
 
     def _validate_freq(self):

pandas/tests/window/test_timeseries_window.py

@@ -1,7 +1,15 @@
 import numpy as np
 import pytest
 
-from pandas import DataFrame, Index, Series, Timestamp, date_range, to_datetime
+from pandas import (
+    DataFrame,
+    Index,
+    MultiIndex,
+    Series,
+    Timestamp,
+    date_range,
+    to_datetime,
+)
 import pandas.util.testing as tm
 
 import pandas.tseries.offsets as offsets
@@ -105,8 +113,16 @@ def test_monotonic_on(self):
         assert df.index.is_monotonic
         df.rolling("2s").sum()
 
-        # non-monotonic
-        df.index = reversed(df.index.tolist())
+    def test_non_monotonic_on(self):
+        # GH 19248
+        df = DataFrame(
+            {"A": date_range("20130101", periods=5, freq="s"), "B": range(5)}
+        )
+        df = df.set_index("A")
+        non_monotonic_index = df.index.to_list()
+        non_monotonic_index[0] = non_monotonic_index[3]
+        df.index = non_monotonic_index
+
         assert not df.index.is_monotonic
 
         with pytest.raises(ValueError):
@@ -690,3 +706,34 @@ def test_rolling_cov_offset(self):
 
         expected2 = ss.rolling(3, min_periods=1).cov()
         tm.assert_series_equal(result, expected2)
+
+    def test_rolling_on_decreasing_index(self):
+        # GH-19248
+        index = [
+            Timestamp("20190101 09:00:00"),
+            Timestamp("20190101 09:00:02"),
+            Timestamp("20190101 09:00:03"),
+            Timestamp("20190101 09:00:05"),
+            Timestamp("20190101 09:00:06"),
+        ]
+
+        df = DataFrame({"column": [3, 4, 4, 2, 1]}, index=reversed(index))
+        result = df.rolling("2s").min()
+        expected = DataFrame(
+            {"column": [3.0, 3.0, 3.0, 2.0, 1.0]}, index=reversed(index)
+        )
+        tm.assert_frame_equal(result, expected)
+
+    def test_rolling_on_multi_index_level(self):
+        # GH-15584
+        df = DataFrame(
+            {"column": range(6)},
+            index=MultiIndex.from_product(
+                [date_range("20190101", periods=3), range(2)], names=["date", "seq"]
+            ),
+        )
+        result = df.rolling("10d", on=df.index.get_level_values("date")).sum()
+        expected = DataFrame(
+            {"column": [0.0, 1.0, 3.0, 6.0, 10.0, 15.0]}, index=df.index
+        )
+        tm.assert_frame_equal(result, expected)