API: Return sparse objects always for cumsum

doc/source/whatsnew/v0.20.0.txt

@@ -70,6 +70,7 @@ Backwards incompatible API changes
 
 
 - ``CParserError`` has been renamed to ``ParserError`` in ``pd.read_csv`` and will be removed in the future (:issue:`12665`)
+- ``SparseArray.cumsum()`` and ``SparseSeries.cumsum()`` will now always return ``SparseArray`` and ``SparseSeries`` respectively (:issue:`12855`)
 
 
 

pandas/sparse/array.py

@@ -620,19 +620,30 @@ def sum(self, axis=0, *args, **kwargs):
 
     def cumsum(self, axis=0, *args, **kwargs):
         """
-        Cumulative sum of values. Preserves locations of NaN values
+        Cumulative sum of non-NA/null values.
+
+        When performing the cumulative summation, any non-NA/null values will
+        be skipped. The resulting SparseArray will preserve the locations of
+        NaN values, but the fill value will be `np.nan` regardless.
+
+        Parameters
+        ----------
+        axis : int or None
+            Axis over which to perform the cumulative summation. If None,
+            perform cumulative summation over flattened array.
 
         Returns
         -------
-        cumsum : Series
+        cumsum : SparseArray
         """
         nv.validate_cumsum(args, kwargs)
 
-        # TODO: gh-12855 - return a SparseArray here
-        if notnull(self.fill_value):
-            return self.to_dense().cumsum()
+        if axis is not None and axis >= self.ndim:  # Mimic ndarray behaviour.
+            raise ValueError("axis(={axis}) out of bounds".format(axis=axis))
+
+        if not self._null_fill_value:
+            return SparseArray(self.to_dense()).cumsum()
 
-        # TODO: what if sp_values contains NaN??
         return SparseArray(self.sp_values.cumsum(), sparse_index=self.sp_index,
                            fill_value=self.fill_value)
 

pandas/sparse/series.py

@@ -630,21 +630,29 @@ def take(self, indices, axis=0, convert=True, *args, **kwargs):
 
     def cumsum(self, axis=0, *args, **kwargs):
         """
-        Cumulative sum of values. Preserves locations of NaN values
+        Cumulative sum of non-NA/null values.
+
+        When performing the cumulative summation, any non-NA/null values will
+        be skipped. The resulting SparseSeries will preserve the locations of
+        NaN values, but the fill value will be `np.nan` regardless.
+
+        Parameters
+        ----------
+        axis : {0}
 
         Returns
         -------
-        cumsum : SparseSeries if `self` has a null `fill_value` and a
-                 generic Series otherwise
+        cumsum : SparseSeries
         """
         nv.validate_cumsum(args, kwargs)
-        new_array = SparseArray.cumsum(self.values)
-        if isinstance(new_array, SparseArray):
-            return self._constructor(
-                new_array, index=self.index,
-                sparse_index=new_array.sp_index).__finalize__(self)
-        # TODO: gh-12855 - return a SparseSeries here
-        return Series(new_array, index=self.index).__finalize__(self)
+        if axis is not None:
+            axis = self._get_axis_number(axis)
+
+        new_array = self.values.cumsum()
+
+        return self._constructor(
+            new_array, index=self.index,
+            sparse_index=new_array.sp_index).__finalize__(self)
 
     @Appender(generic._shared_docs['isnull'])
     def isnull(self):

pandas/sparse/tests/test_array.py

@@ -688,46 +688,57 @@ def test_numpy_sum(self):
                               SparseArray(data), out=out)
 
     def test_cumsum(self):
-        data = np.arange(10).astype(float)
-        out = SparseArray(data).cumsum()
-        expected = SparseArray(data.cumsum())
-        tm.assert_sp_array_equal(out, expected)
+        non_null_data = np.array([1, 2, 3, 4, 5], dtype=float)
+        non_null_expected = SparseArray(non_null_data.cumsum())
 
-        # TODO: gh-12855 - return a SparseArray here
-        data[5] = np.nan
-        out = SparseArray(data, fill_value=2).cumsum()
-        self.assertNotIsInstance(out, SparseArray)
-        tm.assert_numpy_array_equal(out, data.cumsum())
+        null_data = np.array([1, 2, np.nan, 4, 5], dtype=float)
+        null_expected = SparseArray(np.array([1.0, 3.0, np.nan, 7.0, 12.0]))
+
+        for data, expected in [
+            (null_data, null_expected),
+            (non_null_data, non_null_expected)
+        ]:
+            out = SparseArray(data).cumsum()
+            tm.assert_sp_array_equal(out, expected)
+
+            out = SparseArray(data, fill_value=np.nan).cumsum()
+            tm.assert_sp_array_equal(out, expected)
 
-        out = SparseArray(data, fill_value=np.nan).cumsum()
-        expected = SparseArray(np.array([
-            0, 1, 3, 6, 10, np.nan, 16, 23, 31, 40]))
-        tm.assert_sp_array_equal(out, expected)
+            out = SparseArray(data, fill_value=2).cumsum()
+            tm.assert_sp_array_equal(out, expected)
+
+            axis = 1  # SparseArray currently 1-D, so only axis = 0 is valid.
+            msg = "axis\(={axis}\) out of bounds".format(axis=axis)
+            with tm.assertRaisesRegexp(ValueError, msg):
+                SparseArray(data).cumsum(axis=axis)
 
     def test_numpy_cumsum(self):
-        data = np.arange(10).astype(float)
-        out = np.cumsum(SparseArray(data))
-        expected = SparseArray(data.cumsum())
-        tm.assert_sp_array_equal(out, expected)
+        non_null_data = np.array([1, 2, 3, 4, 5], dtype=float)
+        non_null_expected = SparseArray(non_null_data.cumsum())
 
-        # TODO: gh-12855 - return a SparseArray here
-        data[5] = np.nan
-        out = np.cumsum(SparseArray(data, fill_value=2))
-        self.assertNotIsInstance(out, SparseArray)
-        tm.assert_numpy_array_equal(out, data.cumsum())
+        null_data = np.array([1, 2, np.nan, 4, 5], dtype=float)
+        null_expected = SparseArray(np.array([1.0, 3.0, np.nan, 7.0, 12.0]))
 
-        out = np.cumsum(SparseArray(data, fill_value=np.nan))
-        expected = SparseArray(np.array([
-            0, 1, 3, 6, 10, np.nan, 16, 23, 31, 40]))
-        tm.assert_sp_array_equal(out, expected)
+        for data, expected in [
+            (null_data, null_expected),
+            (non_null_data, non_null_expected)
+        ]:
+            out = np.cumsum(SparseArray(data))
+            tm.assert_sp_array_equal(out, expected)
 
-        msg = "the 'dtype' parameter is not supported"
-        tm.assertRaisesRegexp(ValueError, msg, np.cumsum,
-                              SparseArray(data), dtype=np.int64)
+            out = np.cumsum(SparseArray(data, fill_value=np.nan))
+            tm.assert_sp_array_equal(out, expected)
 
-        msg = "the 'out' parameter is not supported"
-        tm.assertRaisesRegexp(ValueError, msg, np.cumsum,
-                              SparseArray(data), out=out)
+            out = np.cumsum(SparseArray(data, fill_value=2))
+            tm.assert_sp_array_equal(out, expected)
+
+            msg = "the 'dtype' parameter is not supported"
+            tm.assertRaisesRegexp(ValueError, msg, np.cumsum,
+                                  SparseArray(data), dtype=np.int64)
+
+            msg = "the 'out' parameter is not supported"
+            tm.assertRaisesRegexp(ValueError, msg, np.cumsum,
+                                  SparseArray(data), out=out)
 
     def test_mean(self):
         data = np.arange(10).astype(float)

pandas/sparse/tests/test_series.py

@@ -1331,21 +1331,22 @@ def test_cumsum(self):
         expected = SparseSeries(self.bseries.to_dense().cumsum())
         tm.assert_sp_series_equal(result, expected)
 
-        # TODO: gh-12855 - return a SparseSeries here
         result = self.zbseries.cumsum()
         expected = self.zbseries.to_dense().cumsum()
-        self.assertNotIsInstance(result, SparseSeries)
         tm.assert_series_equal(result, expected)
 
+        axis = 1  # Series is 1-D, so only axis = 0 is valid.
+        msg = "No axis named {axis}".format(axis=axis)
+        with tm.assertRaisesRegexp(ValueError, msg):
+            self.bseries.cumsum(axis=axis)
+
     def test_numpy_cumsum(self):
         result = np.cumsum(self.bseries)
         expected = SparseSeries(self.bseries.to_dense().cumsum())
         tm.assert_sp_series_equal(result, expected)
 
-        # TODO: gh-12855 - return a SparseSeries here
         result = np.cumsum(self.zbseries)
         expected = self.zbseries.to_dense().cumsum()
-        self.assertNotIsInstance(result, SparseSeries)
         tm.assert_series_equal(result, expected)
 
         msg = "the 'dtype' parameter is not supported"