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.

Sign up for GitHub

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

Jump to bottom

BUG:Floating point accuracy with DatetimeIndex.round (#14440) #15568

Closed
wants to merge 1 commit into from
Closed

BUG:Floating point accuracy with DatetimeIndex.round (#14440) #15568

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions doc/source/whatsnew/v0.20.0.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -651,6 +651,7 @@ Bug Fixes
- Bug in ``Index`` power operations with reversed operands (:issue:`14973`)
- Bug in ``TimedeltaIndex`` addition where overflow was being allowed without error (:issue:`14816`)
- Bug in ``TimedeltaIndex`` raising a ``ValueError`` when boolean indexing with ``loc`` (:issue:`14946`)
- Bug in ``DatetimeIndex.round()`` and ``Timestamp.round()`` floating point accuracy when rounding by milliseconds (:issue: `14440`)
- Bug in ``astype()`` where ``inf`` values were incorrectly converted to integers. Now raises error now with ``astype()`` for Series and DataFrames (:issue:`14265`)
- Bug in ``DataFrame(..).apply(to_numeric)`` when values are of type decimal.Decimal. (:issue:`14827`)
- Bug in ``describe()`` when passing a numpy array which does not contain the median to the ``percentiles`` keyword argument (:issue:`14908`)
Expand Down
11 changes: 11 additions & 0 deletions pandas/tests/indexes/datetimes/test_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -175,6 +175,17 @@ def test_round(self):
tm.assertRaisesRegexp(ValueError, msg, rng.round, freq='M')
tm.assertRaisesRegexp(ValueError, msg, elt.round, freq='M')

# GH 14440
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can you also round to us/ns here as well (which should equal the original)

as an aside, you can now do parametrized tests (but have to move them to separate functions and not class based)

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

So the rounding is happening correctly for microseconds but not with nanoseconds:

In [7]: pd.DatetimeIndex(['2016-10-17 12:00:00.0015']).round('ns')
Out[7]: DatetimeIndex(['2016-10-17 12:00:00.001499904'], dtype='datetime64[ns]', freq=None)

The rounding methodology seems sound. I am unsure if this is a limitation of the date going from int64 to float64 to int64 as this is essentially what is happening:

(Pdb) np.round(np.array([1476705600001500000])/1.).astype('i8')
array([1476705600001499904])

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

yes this is just losing precision - not sure much can be done

we could warn / raise potentially though
would that be useful?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

created a new issue. #15578

index = pd.DatetimeIndex(['2016-10-17 12:00:00.0015'], tz=tz)
result = index.round('ms')
expected = pd.DatetimeIndex(['2016-10-17 12:00:00.002000'], tz=tz)
tm.assert_index_equal(result, expected)

index = pd.DatetimeIndex(['2016-10-17 12:00:00.00149'], tz=tz)
result = index.round('ms')
expected = pd.DatetimeIndex(['2016-10-17 12:00:00.001000'], tz=tz)
tm.assert_index_equal(result, expected)

def test_repeat_range(self):
rng = date_range('1/1/2000', '1/1/2001')

Expand Down
9 changes: 9 additions & 0 deletions pandas/tests/scalar/test_timestamp.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -732,6 +732,15 @@ def test_round(self):
for freq in ['Y', 'M', 'foobar']:
self.assertRaises(ValueError, lambda: dti.round(freq))

# GH 14440
result = pd.Timestamp('2016-10-17 12:00:00.0015').round('ms')
expected = pd.Timestamp('2016-10-17 12:00:00.002000')
self.assertEqual(result, expected)

result = pd.Timestamp('2016-10-17 12:00:00.00149').round('ms')
expected = pd.Timestamp('2016-10-17 12:00:00.001000')
self.assertEqual(result, expected)

def test_class_ops_pytz(self):
tm._skip_if_no_pytz()
from pytz import timezone
Expand Down
2 changes: 1 addition & 1 deletion pandas/tseries/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -83,7 +83,7 @@ def _round(self, freq, rounder):
# round the local times
values = _ensure_datetimelike_to_i8(self)

result = (unit * rounder(values / float(unit))).astype('i8')
result = (unit * rounder(values / float(unit)).astype('i8'))
result = self._maybe_mask_results(result, fill_value=tslib.NaT)
attribs = self._get_attributes_dict()
if 'freq' in attribs:
Expand Down
3 changes: 2 additions & 1 deletion pandas/tslib.pyx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -421,7 +421,8 @@ class Timestamp(_Timestamp):
value = self.tz_localize(None).value
else:
value = self.value
result = Timestamp(unit * rounder(value / float(unit)), unit='ns')
result = (unit * rounder(value / float(unit)).astype('i8'))
result = Timestamp(result, unit='ns')
if self.tz is not None:
result = result.tz_localize(self.tz)
return result
Expand Down