Skip to content

Fix+test timedelta64(nat) ops #23425

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

Merged
merged 6 commits into from
Nov 6, 2018
Merged
Show file tree
Hide file tree
Changes from 3 commits
Commits
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
14 changes: 13 additions & 1 deletion pandas/core/arrays/datetimelike.py
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@
is_object_dtype)
from pandas.core.dtypes.generic import ABCSeries, ABCDataFrame, ABCIndexClass
from pandas.core.dtypes.dtypes import DatetimeTZDtype
from pandas.core.dtypes.missing import isna

import pandas.core.common as com
from pandas.core.algorithms import checked_add_with_arr
Expand Down Expand Up @@ -382,6 +383,12 @@ def _add_timedeltalike_scalar(self, other):
Add a delta of a timedeltalike
return the i8 result view
"""
if isna(other):
Copy link
Contributor

Choose a reason for hiding this comment

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

isn't this pretty much what _nat_new did? I find this is repeating lots of code, why don't you make a method to create the null array instead.

Copy link
Member Author

Choose a reason for hiding this comment

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

_nat_new also did casting+boxing that this doesn't. These two lines aren't repeated often enough to merit a method.

Copy link
Contributor

Choose a reason for hiding this comment

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

ok

# i.e np.timedelta64("NaT"), not recognized by delta_to_nanoseconds
new_values = np.empty(len(self), dtype='i8')
new_values[:] = iNaT
return new_values

inc = delta_to_nanoseconds(other)
new_values = checked_add_with_arr(self.asi8, inc,
arr_mask=self._isnan).view('i8')
Expand Down Expand Up @@ -452,7 +459,7 @@ def _sub_period_array(self, other):
Array of DateOffset objects; nulls represented by NaT
"""
if not is_period_dtype(self):
raise TypeError("cannot subtract {dtype}-dtype to {cls}"
raise TypeError("cannot subtract {dtype}-dtype from {cls}"
.format(dtype=other.dtype,
cls=type(self).__name__))

Expand Down Expand Up @@ -746,6 +753,11 @@ def __rsub__(self, other):
raise TypeError("cannot subtract {cls} from {typ}"
.format(cls=type(self).__name__,
typ=type(other).__name__))
elif is_period_dtype(self) and is_timedelta64_dtype(other):
# TODO: Can we simplify/generalize these cases at all?
raise TypeError("cannot subtract {cls} from {dtype}"
.format(cls=type(self).__name__,
dtype=other.dtype))
return -(self - other)
cls.__rsub__ = __rsub__

Expand Down
9 changes: 8 additions & 1 deletion pandas/core/arrays/period.py
Original file line number Diff line number Diff line change
Expand Up @@ -149,6 +149,8 @@ class PeriodArray(dtl.DatetimeLikeArrayMixin, ExtensionArray):
period_array : Create a new PeriodArray
pandas.PeriodIndex : Immutable Index for period data
"""
# array priority higher than numpy scalars
__array_priority__ = 1000
Copy link
Contributor

Choose a reason for hiding this comment

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

is there a test that needs this? I agree should set it....

Copy link
Member Author

Choose a reason for hiding this comment

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

Yes. IIRC it was timedelta64 - TimedeltaIndex

_attributes = ["freq"]
_typ = "periodarray" # ABCPeriodArray

Expand Down Expand Up @@ -757,7 +759,12 @@ def _add_timedeltalike_scalar(self, other):
assert isinstance(self.freq, Tick) # checked by calling function
assert isinstance(other, (timedelta, np.timedelta64, Tick))

delta = self._check_timedeltalike_freq_compat(other)
if isna(other):
Copy link
Contributor

Choose a reason for hiding this comment

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

I find the need to do this pretty odd, why would't the check routine handle this? I do not like special cases

Copy link
Member Author

Choose a reason for hiding this comment

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

Because the first thing the check function does is call delta_to_nanoseconds, which will raise if we pass np.timedelta64("NaT"). Even if we change that method to avoid calling delta_to_nanoseconds in that case, we'd need to handle a null return value after the check.

I do not like special cases

+1. As Joris mentioned in another thread, a lot of gymnastics seems to be driven by the fact that pd.NaT does double-duty as both Not-A-Time and Not-A-Timedelta.

# special handling for np.timedelta64("NaT"), avoid calling
# _check_timedeltalike_freq_compat as that would raise TypeError
delta = other
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 write this like

if notna(other):
   # your current commnet
    other = self.........

Copy link
Member Author

Choose a reason for hiding this comment

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

Sure

else:
delta = self._check_timedeltalike_freq_compat(other)

# Note: when calling parent class's _add_timedeltalike_scalar,
# it will call delta_to_nanoseconds(delta). Because delta here
Expand Down
2 changes: 1 addition & 1 deletion pandas/core/indexes/base.py
Original file line number Diff line number Diff line change
Expand Up @@ -4706,7 +4706,7 @@ def _evaluate_with_timedelta_like(self, other, op):
'radd', 'rsub']:
raise TypeError("Operation {opname} between {cls} and {other} "
"is invalid".format(opname=op.__name__,
cls=type(self).__name__,
cls=self.dtype,
other=type(other).__name__))

other = Timedelta(other)
Expand Down
19 changes: 19 additions & 0 deletions pandas/tests/arithmetic/test_datetime64.py
Original file line number Diff line number Diff line change
Expand Up @@ -1169,6 +1169,25 @@ def test_dti_isub_timedeltalike(self, tz_naive_fixture, two_hours):
rng -= two_hours
tm.assert_index_equal(rng, expected)

def test_dt64arr_add_sub_td64_nat(self, box, tz_naive_fixture):
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 parameterize other with pd.NaT as well here (I bet we already have a test for that, can you consolidate)

Copy link
Member Author

Choose a reason for hiding this comment

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

Not really; pd.NaT behavior is pretty starkly different from the behavior we’re testing here.

That said, there are several rounds of de duplication and parameterization coming up

Copy link
Contributor

Choose a reason for hiding this comment

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

how is it different at all?

Copy link
Member Author

Choose a reason for hiding this comment

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

Because this is a timedelta64 and pd.NaT is in most cases datetime-like. The "most cases" part makes it especially hairy.

# GH#23320 special handling for timedelta64("NaT")
tz = tz_naive_fixture
dti = pd.date_range("1994-04-01", periods=9, tz=tz, freq="QS")
other = np.timedelta64("NaT")
expected = pd.DatetimeIndex(["NaT"] * 9, tz=tz)

obj = tm.box_expected(dti, box)
expected = tm.box_expected(expected, box)

result = obj + other
tm.assert_equal(result, expected)
result = other + obj
tm.assert_equal(result, expected)
result = obj - other
tm.assert_equal(result, expected)
with pytest.raises(TypeError):
other - obj
Copy link
Member

Choose a reason for hiding this comment

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

I see a lot of error message changes. Are they tested anywhere with assert_raises_regex?

Copy link
Member Author

Choose a reason for hiding this comment

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

The way these tests are parametrized, the messages are all over the place.


# -------------------------------------------------------------
# Binary operations DatetimeIndex and TimedeltaIndex/array
def test_dti_add_tdi(self, tz_naive_fixture):
Expand Down
20 changes: 19 additions & 1 deletion pandas/tests/arithmetic/test_period.py
Original file line number Diff line number Diff line change
Expand Up @@ -419,7 +419,7 @@ def test_pi_add_sub_td64_array_non_tick_raises(self):

with pytest.raises(period.IncompatibleFrequency):
rng - tdarr
with pytest.raises(period.IncompatibleFrequency):
with pytest.raises(TypeError):
tdarr - rng

def test_pi_add_sub_td64_array_tick(self):
Expand Down Expand Up @@ -785,6 +785,24 @@ def test_pi_add_sub_timedeltalike_freq_mismatch_monthly(self,
with tm.assert_raises_regex(period.IncompatibleFrequency, msg):
rng -= other

def test_parr_add_sub_td64_nat(self, box):
Copy link
Contributor

Choose a reason for hiding this comment

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

same as above

# GH#23320 special handling for timedelta64("NaT")
pi = pd.period_range("1994-04-01", periods=9, freq="19D")
other = np.timedelta64("NaT")
expected = pd.PeriodIndex(["NaT"] * 9, freq="19D")

obj = tm.box_expected(pi, box)
expected = tm.box_expected(expected, box)

result = obj + other
tm.assert_equal(result, expected)
result = other + obj
tm.assert_equal(result, expected)
result = obj - other
tm.assert_equal(result, expected)
with pytest.raises(TypeError):
other - obj


class TestPeriodSeriesArithmetic(object):
def test_ops_series_timedelta(self):
Expand Down
18 changes: 18 additions & 0 deletions pandas/tests/arithmetic/test_timedelta64.py
Original file line number Diff line number Diff line change
Expand Up @@ -735,6 +735,24 @@ def test_td64arr_add_sub_tdi(self, box_df_broadcast_failure, names):
else:
assert result.dtypes[0] == 'timedelta64[ns]'

def test_td64arr_add_sub_td64_nat(self, box):
# GH#23320 special handling for timedelta64("NaT")
Copy link
Contributor

Choose a reason for hiding this comment

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

same

tdi = pd.TimedeltaIndex([NaT, Timedelta('1s')])
other = np.timedelta64("NaT")
expected = pd.TimedeltaIndex(["NaT"] * 2)

obj = tm.box_expected(tdi, box)
expected = tm.box_expected(expected, box)

result = obj + other
tm.assert_equal(result, expected)
result = other + obj
tm.assert_equal(result, expected)
result = obj - other
tm.assert_equal(result, expected)
result = other - obj
tm.assert_equal(result, expected)

def test_td64arr_sub_NaT(self, box):
# GH#18808
ser = Series([NaT, Timedelta('1s')])
Expand Down