Behaviour of comparisons between pandas.Timestamp and datetime.date objects changes in Pandas v2.0

[matt-graham]

Pandas v2.0 changes the behaviour of equality comparisons of pandas.Timestamp and datetime.date objects to always compare as unequal (pandas-dev/pandas#36131). For example

import datetime, pandas
print(pandas.Timestamp(2010, 1, 1) == datetime.date(2010, 1, 1))

outputs True in Pandas v1.2.2 and False in v2.0.0.

This causes the assert in tests/test_control_order_of_events.py::test_control_of_ordering_in_the_day

TLOmodel/tests/test_control_order_of_events.py

Lines 92 to 97 in 5655119

	events['date'] = pd.to_datetime(events['date']).dt.date
	order_on_day_one = tuple(events.loc[events['date'] == Date(2010, 1, 1), 'id'])
	assert order_on_day_one == ("EventForStartOfDay",
	"EventForMiddleOfDay",
	"EventForSecondToLastAtEndOfDay",
	"EventForEndOfDay")

to fail when running with Pandas v2.0.0 as the earlier comparison with a Date object (corresponding to a pandas.Timestamp) no longer behaves as expected.

[matt-graham]

Found another instance of this causing a problem in tests/test_healthcareseeking.py::test_force_healthcare_seeking

TLOmodel/tests/test_healthcareseeking.py

Lines 911 to 912 in 5655119

	evs = [x[1].TREATMENT_ID for x in
	sim.modules['HealthSystem'].find_events_for_person(0) if x[0].date() == start_date]

The x[0] tuple entry here is a pd.Timestamp object already so just removing the .date() call which convert is to a datetime.date object appears to fix things. @tbhallett can you remember why you added this conversion in this first place?

[tbhallett]

Found another instance of this causing a problem in tests/test_healthcareseeking.py::test_force_healthcare_seeking

TLOmodel/tests/test_healthcareseeking.py

Lines 911 to 912 in 5655119

	evs = [x[1].TREATMENT_ID for x in
	sim.modules['HealthSystem'].find_events_for_person(0) if x[0].date() == start_date]

The x[0] tuple entry here is a pd.Timestamp object already so just removing the .date() call which convert is to a datetime.date object appears to fix things. @tbhallett can you remember why you added this conversion in this first place?

I can't! But the changes seem good to me. Thanks Matt.