Backport PR pandas-dev#38909: BUG: Fixed regression in rolling.skew and rolling.kurt modifying object

Author: phofl

doc/source/whatsnew/v1.2.1.rst

@@ -37,7 +37,7 @@ Fixed regressions
 - Fixed regression in :meth:`.GroupBy.sem` where the presence of non-numeric columns would cause an error instead of being dropped (:issue:`38774`)
 - Fixed regression in :func:`read_excel` with non-rawbyte file handles (:issue:`38788`)
 - Bug in :meth:`read_csv` with ``float_precision="high"`` caused segfault or wrong parsing of long exponent strings. This resulted in a regression in some cases as the default for ``float_precision`` was changed in pandas 1.2.0 (:issue:`38753`)
--
+- Fixed regression in :meth:`Rolling.skew` and :meth:`Rolling.kurt` modifying the object inplace (:issue:`38908`)
 
 .. ---------------------------------------------------------------------------
 

pandas/_libs/window/aggregations.pyx

@@ -523,7 +523,7 @@ def roll_skew(ndarray[float64_t] values, ndarray[int64_t] start,
         float64_t x = 0, xx = 0, xxx = 0
         int64_t nobs = 0, i, j, N = len(values), nobs_mean = 0
         int64_t s, e
-        ndarray[float64_t] output, mean_array
+        ndarray[float64_t] output, mean_array, values_copy
         bint is_monotonic_increasing_bounds
 
     minp = max(minp, 3)
@@ -532,10 +532,11 @@ def roll_skew(ndarray[float64_t] values, ndarray[int64_t] start,
     )
     output = np.empty(N, dtype=float)
     min_val = np.nanmin(values)
+    values_copy = np.copy(values)
 
     with nogil:
         for i in range(0, N):
-            val = values[i]
+            val = values_copy[i]
             if notnan(val):
                 nobs_mean += 1
                 sum_val += val
@@ -544,7 +545,7 @@ def roll_skew(ndarray[float64_t] values, ndarray[int64_t] start,
         if min_val - mean_val > -1e5:
             mean_val = round(mean_val)
             for i in range(0, N):
-                values[i] = values[i] - mean_val
+                values_copy[i] = values_copy[i] - mean_val
 
         for i in range(0, N):
 
@@ -556,7 +557,7 @@ def roll_skew(ndarray[float64_t] values, ndarray[int64_t] start,
             if i == 0 or not is_monotonic_increasing_bounds:
 
                 for j in range(s, e):
-                    val = values[j]
+                    val = values_copy[j]
                     add_skew(val, &nobs, &x, &xx, &xxx, &compensation_x_add,
                              &compensation_xx_add, &compensation_xxx_add)
 
@@ -566,13 +567,13 @@ def roll_skew(ndarray[float64_t] values, ndarray[int64_t] start,
                 # and removed
                 # calculate deletes
                 for j in range(start[i - 1], s):
-                    val = values[j]
+                    val = values_copy[j]
                     remove_skew(val, &nobs, &x, &xx, &xxx, &compensation_x_remove,
                                 &compensation_xx_remove, &compensation_xxx_remove)
 
                 # calculate adds
                 for j in range(end[i - 1], e):
-                    val = values[j]
+                    val = values_copy[j]
                     add_skew(val, &nobs, &x, &xx, &xxx, &compensation_x_add,
                              &compensation_xx_add, &compensation_xxx_add)
 
@@ -703,19 +704,20 @@ def roll_kurt(ndarray[float64_t] values, ndarray[int64_t] start,
         float64_t compensation_x_remove = 0, compensation_x_add = 0
         float64_t x = 0, xx = 0, xxx = 0, xxxx = 0
         int64_t nobs = 0, i, j, s, e, N = len(values), nobs_mean = 0
-        ndarray[float64_t] output
+        ndarray[float64_t] output, values_copy
         bint is_monotonic_increasing_bounds
 
     minp = max(minp, 4)
     is_monotonic_increasing_bounds = is_monotonic_increasing_start_end_bounds(
         start, end
     )
     output = np.empty(N, dtype=float)
+    values_copy = np.copy(values)
     min_val = np.nanmin(values)
 
     with nogil:
         for i in range(0, N):
-            val = values[i]
+            val = values_copy[i]
             if notnan(val):
                 nobs_mean += 1
                 sum_val += val
@@ -724,7 +726,7 @@ def roll_kurt(ndarray[float64_t] values, ndarray[int64_t] start,
         if min_val - mean_val > -1e4:
             mean_val = round(mean_val)
             for i in range(0, N):
-                values[i] = values[i] - mean_val
+                values_copy[i] = values_copy[i] - mean_val
 
         for i in range(0, N):
 
@@ -736,7 +738,7 @@ def roll_kurt(ndarray[float64_t] values, ndarray[int64_t] start,
             if i == 0 or not is_monotonic_increasing_bounds:
 
                 for j in range(s, e):
-                    add_kurt(values[j], &nobs, &x, &xx, &xxx, &xxxx,
+                    add_kurt(values_copy[j], &nobs, &x, &xx, &xxx, &xxxx,
                              &compensation_x_add, &compensation_xx_add,
                              &compensation_xxx_add, &compensation_xxxx_add)
 
@@ -746,13 +748,13 @@ def roll_kurt(ndarray[float64_t] values, ndarray[int64_t] start,
                 # and removed
                 # calculate deletes
                 for j in range(start[i - 1], s):
-                    remove_kurt(values[j], &nobs, &x, &xx, &xxx, &xxxx,
+                    remove_kurt(values_copy[j], &nobs, &x, &xx, &xxx, &xxxx,
                                 &compensation_x_remove, &compensation_xx_remove,
                                 &compensation_xxx_remove, &compensation_xxxx_remove)
 
                 # calculate adds
                 for j in range(end[i - 1], e):
-                    add_kurt(values[j], &nobs, &x, &xx, &xxx, &xxxx,
+                    add_kurt(values_copy[j], &nobs, &x, &xx, &xxx, &xxxx,
                              &compensation_x_add, &compensation_xx_add,
                              &compensation_xxx_add, &compensation_xxxx_add)
 

pandas/tests/window/test_rolling.py

@@ -1102,11 +1102,13 @@ def test_groupby_rolling_nan_included():
 
 @pytest.mark.parametrize("method", ["skew", "kurt"])
 def test_rolling_skew_kurt_numerical_stability(method):
-    # GH: 6929
-    s = Series(np.random.rand(10))
-    expected = getattr(s.rolling(3), method)()
-    s = s + 50000
-    result = getattr(s.rolling(3), method)()
+    # GH#6929
+    ser = Series(np.random.rand(10))
+    ser_copy = ser.copy()
+    expected = getattr(ser.rolling(3), method)()
+    tm.assert_series_equal(ser, ser_copy)
+    ser = ser + 50000
+    result = getattr(ser.rolling(3), method)()
     tm.assert_series_equal(result, expected)