[REVIEW] Add support for BaseIndexer in Rolling APIs

python/cudf/cudf/core/column/numerical.py

@@ -375,6 +375,9 @@ def fillna(
         else:
             col = self
 
+        if col.null_count == 0:
+            return col
+
         if method is not None:
             return super(NumericalColumn, col).fillna(fill_value, method)
 

python/cudf/cudf/core/window/rolling.py

@@ -4,6 +4,7 @@
 
 import numba
 import pandas as pd
+from pandas.api.indexers import BaseIndexer
 
 import cudf
 from cudf import _lib as libcudf
@@ -20,14 +21,16 @@ class Rolling(GetAttrGetItemMixin):
 
     Parameters
     ----------
-    window : int or offset
+    window : int, offset or a BaseIndexer subclass
         Size of the window, i.e., the number of observations used
         to calculate the statistic.
         For datetime indexes, an offset can be provided instead
         of an int. The offset must be convertible to a timedelta.
         As opposed to a fixed window size, each window will be
         sized to accommodate observations within the time period
         specified by the offset.
+        If a BaseIndexer subclass is passed, calculates the window
+        boundaries based on the defined ``get_window_bounds`` method.
     min_periods : int, optional
         The minimum number of observations in the window that are
         required to be non-null, so that the result is non-null.
@@ -195,26 +198,46 @@ def __getitem__(self, arg):
         )
 
     def _apply_agg_series(self, sr, agg_name):
+        source_column = sr._column
+        min_periods = self.min_periods or 1
         if isinstance(self.window, int):
-            result_col = libcudf.rolling.rolling(
-                sr._column,
-                None,
-                None,
-                self.window,
-                self.min_periods,
-                self.center,
-                agg_name,
+            preceding_window = None
+            following_window = None
+            window = self.window
+        elif isinstance(self.window, BaseIndexer):
+            start, end = self.window.get_window_bounds(
+                num_values=len(self.obj),
+                min_periods=self.min_periods,
+                center=self.center,
+                closed=None,
             )
+            start = as_column(start, dtype="int32")
+            end = as_column(end, dtype="int32")
+
+            idx = cudf.core.column.arange(len(start))
+            preceding_window = (idx - start + cudf.Scalar(1, "int32")).astype(
+                "int32"
+            )
+            following_window = (end - idx - cudf.Scalar(1, "int32")).astype(
+                "int32"
+            )
+            window = None
         else:
-            result_col = libcudf.rolling.rolling(
-                sr._column,
-                as_column(self.window),
-                column.full(self.window.size, 0, dtype=self.window.dtype),
-                None,
-                self.min_periods,
-                self.center,
-                agg_name,
+            preceding_window = as_column(self.window)
+            following_window = column.full(
+                self.window.size, 0, dtype=self.window.dtype
             )
+            window = None
+
+        result_col = libcudf.rolling.rolling(
+            source_column=source_column,
+            pre_column_window=preceding_window,
+            fwd_column_window=following_window,
+            window=window,
+            min_periods=min_periods,
+            center=self.center,
+            op=agg_name,
+        )
         return sr._from_data({sr.name: result_col}, sr._index)
 
     def _apply_agg_dataframe(self, df, agg_name):
@@ -305,15 +328,17 @@ def _normalize(self):
             if self.min_periods is None:
                 min_periods = window
         else:
-            if isinstance(window, numba.cuda.devicearray.DeviceNDArray):
-                # window is a device_array of window sizes
+            if isinstance(
+                window, (numba.cuda.devicearray.DeviceNDArray, BaseIndexer)
+            ):
+                # window is a device_array of window sizes or BaseIndexer
                 self.window = window
                 self.min_periods = min_periods
                 return
 
             if not isinstance(self.obj.index, cudf.core.index.DatetimeIndex):
                 raise ValueError(
-                    "window must be an integer for " "non datetime index"
+                    "window must be an integer for non datetime index"
                 )
 
             self._time_window = True
@@ -326,7 +351,7 @@ def _normalize(self):
                 window = window.to_timedelta64()
             except ValueError as e:
                 raise ValueError(
-                    "window must be integer or " "convertible to a timedelta"
+                    "window must be integer or convertible to a timedelta"
                 ) from e
             if self.min_periods is None:
                 min_periods = 1

python/cudf/cudf/tests/test_replace.py

@@ -657,6 +657,7 @@ def test_fillna_method_fixed_width_non_num(data, container, method, inplace):
         pd.DataFrame(
             {"a": [1, 2, None], "b": [None, None, 5]}, index=["a", "p", "z"]
         ),
+        pd.DataFrame({"a": [1, 2, 3]}),
     ],
 )
 @pytest.mark.parametrize(
@@ -671,6 +672,7 @@ def test_fillna_method_fixed_width_non_num(data, container, method, inplace):
         {"b": pd.Series([11, 22, 33], index=["a", "p", "z"])},
         {"a": 5, "b": pd.Series([3, 4, 5], index=["a", "p", "z"])},
         {"c": 100},
+        np.nan,
     ],
 )
 @pytest.mark.parametrize("inplace", [True, False])

python/cudf/cudf/tests/test_rolling.py

@@ -369,3 +369,51 @@ def test_rolling_groupby_offset(agg, window_size):
     )
     got = getattr(gdf.groupby("group").rolling(window_size), agg)().fillna(-1)
     assert_eq(expect, got, check_dtype=False)
+
+
+def test_rolling_custom_index_support():
+    from pandas.api.indexers import BaseIndexer
+
+    class CustomIndexer(BaseIndexer):
+        def get_window_bounds(self, num_values, min_periods, center, closed):
+            start = np.empty(num_values, dtype=np.int64)
+            end = np.empty(num_values, dtype=np.int64)
+
+            for i in range(num_values):
+                if self.use_expanding[i]:
+                    start[i] = 0
+                    end[i] = i + 1
+                else:
+                    start[i] = i
+                    end[i] = i + self.window_size
+
+            return start, end
+
+    use_expanding = [True, False, True, False, True]
+    indexer = CustomIndexer(window_size=1, use_expanding=use_expanding)
+
+    df = pd.DataFrame({"values": range(5)})
+    gdf = cudf.from_pandas(df)
+
+    expected = df.rolling(window=indexer).sum()
+    actual = gdf.rolling(window=indexer).sum()
+
+    assert_eq(expected, actual, check_dtype=False)
+
+
+@pytest.mark.parametrize(
+    "indexer",
+    [
+        pd.api.indexers.FixedForwardWindowIndexer(window_size=2),
+        pd.core.window.indexers.ExpandingIndexer(),
+        pd.core.window.indexers.FixedWindowIndexer(window_size=3),
+    ],
+)
+def test_rolling_indexer_support(indexer):
+    df = pd.DataFrame({"B": [0, 1, 2, np.nan, 4]})
+    gdf = cudf.from_pandas(df)
+
+    expected = df.rolling(window=indexer, min_periods=2).sum()
+    actual = gdf.rolling(window=indexer, min_periods=2).sum()
+
+    assert_eq(expected, actual)