Backport PR #43467: Make group_mean compatible with NaT (#43630)

Co-authored-by: Alexey Györi <[email protected]>

Author: meeseeksmachine

doc/source/whatsnew/v1.3.4.rst

@@ -27,8 +27,7 @@ Fixed regressions
 
 Bug fixes
 ~~~~~~~~~
--
--
+- Fixed bug in :meth:`.GroupBy.mean` with datetimelike values including ``NaT`` values returning incorrect results (:issue`:43132`)
 
 .. ---------------------------------------------------------------------------
 

pandas/_libs/groupby.pyi

@@ -74,7 +74,10 @@ def group_mean(
     counts: np.ndarray,  # int64_t[::1]
     values: np.ndarray,  # ndarray[floating, ndim=2]
     labels: np.ndarray,  # const intp_t[:]
-    min_count: int = ...,
+    min_count: int = ...,  # Py_ssize_t
+    is_datetimelike: bool = ...,  # bint
+    mask: np.ndarray | None = ...,
+    result_mask: np.ndarray | None = ...,
 ) -> None: ...
 def group_ohlc(
     out: np.ndarray,  # floating[:, ::1]

pandas/_libs/groupby.pyx

@@ -669,10 +669,45 @@ def group_mean(floating[:, ::1] out,
                int64_t[::1] counts,
                ndarray[floating, ndim=2] values,
                const intp_t[::1] labels,
-               Py_ssize_t min_count=-1) -> None:
+               Py_ssize_t min_count=-1,
+               bint is_datetimelike=False,
+               const uint8_t[:, ::1] mask=None,
+               uint8_t[:, ::1] result_mask=None
+               ) -> None:
+    """
+    Compute the mean per label given a label assignment for each value.
+    NaN values are ignored.
+
+    Parameters
+    ----------
+    out : np.ndarray[floating]
+        Values into which this method will write its results.
+    counts : np.ndarray[int64]
+        A zeroed array of the same shape as labels,
+        populated by group sizes during algorithm.
+    values : np.ndarray[floating]
+        2-d array of the values to find the mean of.
+    labels : np.ndarray[np.intp]
+        Array containing unique label for each group, with its
+        ordering matching up to the corresponding record in `values`.
+    min_count : Py_ssize_t
+        Only used in add and prod. Always -1.
+    is_datetimelike : bool
+        True if `values` contains datetime-like entries.
+    mask : ndarray[bool, ndim=2], optional
+        Not used.
+    result_mask : ndarray[bool, ndim=2], optional
+        Not used.
+
+    Notes
+    -----
+    This method modifies the `out` parameter rather than returning an object.
+    `counts` is modified to hold group sizes
+    """
+
     cdef:
         Py_ssize_t i, j, N, K, lab, ncounts = len(counts)
-        floating val, count, y, t
+        floating val, count, y, t, nan_val
         floating[:, ::1] sumx, compensation
         int64_t[:, ::1] nobs
         Py_ssize_t len_values = len(values), len_labels = len(labels)
@@ -682,12 +717,13 @@ def group_mean(floating[:, ::1] out,
     if len_values != len_labels:
         raise ValueError("len(index) != len(labels)")
 
-    nobs = np.zeros((<object>out).shape, dtype=np.int64)
     # the below is equivalent to `np.zeros_like(out)` but faster
+    nobs = np.zeros((<object>out).shape, dtype=np.int64)
     sumx = np.zeros((<object>out).shape, dtype=(<object>out).base.dtype)
     compensation = np.zeros((<object>out).shape, dtype=(<object>out).base.dtype)
 
     N, K = (<object>values).shape
+    nan_val = NPY_NAT if is_datetimelike else NAN
 
     with nogil:
         for i in range(N):
@@ -699,7 +735,7 @@ def group_mean(floating[:, ::1] out,
             for j in range(K):
                 val = values[i, j]
                 # not nan
-                if val == val:
+                if val == val and not (is_datetimelike and val == NPY_NAT):
                     nobs[lab, j] += 1
                     y = val - compensation[lab, j]
                     t = sumx[lab, j] + y
@@ -710,7 +746,7 @@ def group_mean(floating[:, ::1] out,
             for j in range(K):
                 count = nobs[i, j]
                 if nobs[i, j] == 0:
-                    out[i, j] = NAN
+                    out[i, j] = nan_val
                 else:
                     out[i, j] = sumx[i, j] / count
 

pandas/core/groupby/ops.py

@@ -544,7 +544,7 @@ def _call_cython_op(
         result = maybe_fill(np.empty(out_shape, dtype=out_dtype))
         if self.kind == "aggregate":
             counts = np.zeros(ngroups, dtype=np.int64)
-            if self.how in ["min", "max"]:
+            if self.how in ["min", "max", "mean"]:
                 func(
                     result,
                     counts,

pandas/tests/groupby/aggregate/test_aggregate.py

@@ -20,6 +20,7 @@
     MultiIndex,
     Series,
     concat,
+    to_datetime,
 )
 import pandas._testing as tm
 from pandas.core.base import SpecificationError
@@ -66,7 +67,6 @@ def test_agg_ser_multi_key(df):
 
 
 def test_groupby_aggregation_mixed_dtype():
-
     # GH 6212
     expected = DataFrame(
         {
@@ -1274,3 +1274,35 @@ def func(ser):
 
     expected = DataFrame([[1.0]], index=[1])
     tm.assert_frame_equal(res, expected)
+
+
+def test_group_mean_timedelta_nat():
+    # GH43132
+    data = Series(["1 day", "3 days", "NaT"], dtype="timedelta64[ns]")
+    expected = Series(["2 days"], dtype="timedelta64[ns]")
+
+    result = data.groupby([0, 0, 0]).mean()
+
+    tm.assert_series_equal(result, expected)
+
+
+@pytest.mark.parametrize(
+    "input_data, expected_output",
+    [
+        (  # no timezone
+            ["2021-01-01T00:00", "NaT", "2021-01-01T02:00"],
+            ["2021-01-01T01:00"],
+        ),
+        (  # timezone
+            ["2021-01-01T00:00-0100", "NaT", "2021-01-01T02:00-0100"],
+            ["2021-01-01T01:00-0100"],
+        ),
+    ],
+)
+def test_group_mean_datetime64_nat(input_data, expected_output):
+    # GH43132
+    data = to_datetime(Series(input_data))
+    expected = to_datetime(Series(expected_output))
+
+    result = data.groupby([0, 0, 0]).mean()
+    tm.assert_series_equal(result, expected)

pandas/tests/groupby/test_libgroupby.py

@@ -1,9 +1,11 @@
 import numpy as np
+import pytest
 
 from pandas._libs import groupby as libgroupby
 from pandas._libs.groupby import (
     group_cumprod_float64,
     group_cumsum,
+    group_mean,
     group_var,
 )
 
@@ -234,3 +236,51 @@ def test_cython_group_transform_algos():
         ]
     )
     tm.assert_numpy_array_equal(actual[:, 0].view("m8[ns]"), expected)
+
+
+def test_cython_group_mean_datetimelike():
+    actual = np.zeros(shape=(1, 1), dtype="float64")
+    counts = np.array([0], dtype="int64")
+    data = (
+        np.array(
+            [np.timedelta64(2, "ns"), np.timedelta64(4, "ns"), np.timedelta64("NaT")],
+            dtype="m8[ns]",
+        )[:, None]
+        .view("int64")
+        .astype("float64")
+    )
+    labels = np.zeros(len(data), dtype=np.intp)
+
+    group_mean(actual, counts, data, labels, is_datetimelike=True)
+
+    tm.assert_numpy_array_equal(actual[:, 0], np.array([3], dtype="float64"))
+
+
+def test_cython_group_mean_wrong_min_count():
+    actual = np.zeros(shape=(1, 1), dtype="float64")
+    counts = np.zeros(1, dtype="int64")
+    data = np.zeros(1, dtype="float64")[:, None]
+    labels = np.zeros(1, dtype=np.intp)
+
+    with pytest.raises(AssertionError, match="min_count"):
+        group_mean(actual, counts, data, labels, is_datetimelike=True, min_count=0)
+
+
+def test_cython_group_mean_not_datetimelike_but_has_NaT_values():
+    actual = np.zeros(shape=(1, 1), dtype="float64")
+    counts = np.array([0], dtype="int64")
+    data = (
+        np.array(
+            [np.timedelta64("NaT"), np.timedelta64("NaT")],
+            dtype="m8[ns]",
+        )[:, None]
+        .view("int64")
+        .astype("float64")
+    )
+    labels = np.zeros(len(data), dtype=np.intp)
+
+    group_mean(actual, counts, data, labels, is_datetimelike=False)
+
+    tm.assert_numpy_array_equal(
+        actual[:, 0], np.array(np.divide(np.add(data[0], data[1]), 2), dtype="float64")
+    )