Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

BUG: .rolling() returns incorrect values when ts index is not nano seconds #55173

Merged
merged 20 commits into from Oct 26, 2023
Merged
Show file tree
Hide file tree
Changes from 13 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
2 changes: 2 additions & 0 deletions doc/source/whatsnew/v2.1.2.rst
Expand Up @@ -16,6 +16,8 @@ Fixed regressions
- Fixed bug in :meth:`DataFrame.resample` where bin edges were not correct for :class:`~pandas.tseries.offsets.MonthBegin` (:issue:`55271`)
- Fixed bug where PDEP-6 warning about setting an item of an incompatible dtype was being shown when creating a new conditional column (:issue:`55025`)
- Fixed regression in :meth:`DataFrame.join` where result has missing values and dtype is arrow backed string (:issue:`55348`)
- Fixed bug in :class:`pandas.core.window.Rolling` where non-nanosecond ``on`` would produce incorrect results (:issue:`55026`, :issue:`55106`, :issue:`55299`)
-

.. ---------------------------------------------------------------------------
.. _whatsnew_212.bug_fixes:
Expand Down
10 changes: 10 additions & 0 deletions pandas/core/window/rolling.py
Expand Up @@ -112,6 +112,8 @@
from pandas.core.generic import NDFrame
from pandas.core.groupby.ops import BaseGrouper

from pandas.core.arrays.datetimelike import dtype_to_unit


class BaseWindow(SelectionMixin):
"""Provides utilities for performing windowing operations."""
Expand Down Expand Up @@ -1889,6 +1891,14 @@ def _validate(self):
else:
self._win_freq_i8 = freq.nanos

unit = dtype_to_unit(self._on.dtype)
if unit == "us":
self._win_freq_i8 /= 1e3
hkhojasteh marked this conversation as resolved.
Show resolved Hide resolved
elif unit == "ms":
self._win_freq_i8 /= 1e6
elif unit == "s":
self._win_freq_i8 /= 1e9

# min_periods must be an integer
if self.min_periods is None:
self.min_periods = 1
Expand Down
32 changes: 32 additions & 0 deletions pandas/tests/window/test_rolling.py
Expand Up @@ -1950,3 +1950,35 @@ def test_numeric_only_corr_cov_series(kernel, use_arg, numeric_only, dtype):
op2 = getattr(rolling2, kernel)
expected = op2(*arg2, numeric_only=numeric_only)
tm.assert_series_equal(result, expected)


@pytest.mark.parametrize("unit", ["s", "ms", "us", "ns"])
@pytest.mark.parametrize("tz", [None, "UTC", "Europe/Prague"])
def test_rolling_timedelta_window_non_nanoseconds(unit, tz):
# Test Sum, GH#55106
df_time = DataFrame(
{"A": range(5)}, index=date_range("2013-01-01", freq="1s", periods=5, tz=tz)
)
sum_in_nanosecs = df_time.rolling("1s").sum()
# microseconds / milliseconds should not break the correct rolling
df_time.index = df_time.index.as_unit(unit)
sum_in_microsecs = df_time.rolling("1s").sum()
sum_in_microsecs.index = sum_in_microsecs.index.as_unit("ns")
tm.assert_frame_equal(sum_in_nanosecs, sum_in_microsecs)

# Test max, GH#55026
ref_dates = date_range("2023-01-01", "2023-01-10", unit="ns", tz=tz)
ref_series = Series(0, index=ref_dates)
ref_series.iloc[0] = 1
ref_max_series = ref_series.rolling(Timedelta(days=4)).max()

dates = date_range("2023-01-01", "2023-01-10", unit=unit, tz=tz)
series = Series(0, index=dates)
series.iloc[0] = 1
max_series = series.rolling(Timedelta(days=4)).max()

ref_df = DataFrame(ref_max_series)
df = DataFrame(max_series)
df.index = df.index.as_unit("ns")

tm.assert_frame_equal(ref_df, df)