Repository navigation
TST: Test failures with pandas 1.3.0 #11918
Description
pandas 1.3.0 was released 3 hours ago (as of writing this): https://github.com/pandas-dev/pandas/releases/tag/v1.3.0
It is causing the following test failures. I think we can just update the test results, but I want to make sure that subpackage maintainers are fine with this plan. Doctest is trickier but we can just require pandas>=1.3 for those?
________________ [doctest] astropy.table.table.Table.to_pandas _________________
3572
3573 >>> q = [1, 2] * u.m
3574 >>> tm = Time([1998, 2002], format='jyear')
3575 >>> sc = SkyCoord([5, 6], [7, 8], unit='deg')
3576 >>> dt = TimeDelta([3, 200] * u.s)
3577
3578 >>> t = QTable([q, tm, sc, dt], names=['q', 'tm', 'sc', 'dt'])
3579
3580 >>> df = t.to_pandas(index='tm')
3581 >>> with pd.option_context('display.max_columns', 20):
Differences (unified diff with -expected +actual):
@@ -1,4 +1,3 @@
q sc.ra sc.dec dt
-tm
1998-01-01 1.0 5.0 7.0 0 days 00:00:03
2002-01-01 2.0 6.0 8.0 0 days 00:03:20
/Users/runner/work/astropy/astropy/.tox/py37-test-alldeps/lib/python3.7/site-packages/astropy/table/table.py:3581: DocTestFailure
______________________ TestPandas.test_from_pandas_index _______________________
self = <astropy.table.tests.test_table.TestPandas object at 0x1346a5390>
def test_from_pandas_index(self):
tm = Time([1998, 2002], format='jyear')
x = [1, 2]
t = table.Table([tm, x], names=['tm', 'x'])
tp = t.to_pandas(index='tm')
t2 = table.Table.from_pandas(tp)
assert t2.colnames == ['x']
t2 = table.Table.from_pandas(tp, index=True)
> assert t2.colnames == ['tm', 'x']
E AssertionError: assert ['index', 'x'] == ['tm', 'x']
E At index 0 diff: 'index' != 'tm'
E Use -v to get the full diff
../../.tox/py37-test-alldeps/lib/python3.7/site-packages/astropy/table/tests/test_table.py:1961: AssertionError
_____________________________ [doctest] pandas.rst _____________________________
059 2015-07-06T00:00:00.000000000 3
060
061 Converting to :class:`~pandas.DataFrame` can also be done with
062 :meth:`~astropy.timeseries.TimeSeries.to_pandas`:
063
064 .. doctest-requires:: pandas
065
066 >>> ts['b'] = [1.2, 3.4, 5.4]
067 >>> df_new = ts.to_pandas()
068 >>> df_new
Differences (unified diff with -expected +actual):
@@ -1,4 +1,3 @@
a b
-time
2015-07-04 1 1.2
2015-07-05 2 3.4
/Users/runner/work/astropy/astropy/docs/timeseries/pandas.rst:68: DocTestFailure
Example log with pandas 1.3.0 (failed): https://github.com/astropy/astropy/runs/2973199416?check_suite_focus=true
Example log with pandas 1.2.5 (passed): https://github.com/astropy/astropy/runs/2968371046?check_suite_focus=true
Activity
mhvk commentedon Jul 2, 2021
It seems strange that the name of the index column is lost, so not sure just updating the test is OK. But I don't understand pandas well enough to know...
pllim commentedon Jul 2, 2021
Yeah... Perhaps @taldcroft could advise or if not, maybe ping someone from
pandasto make sure this change is intentional.pllim commentedon Jul 2, 2021
I wanted to test this locally but
pip install pandas -Urefuses to upgrade from 1.2.x to 1.3 on Python 3.9. 🤯taldcroft commentedon Jul 3, 2021
Hopefully fixed by #11921.
taldcroft commentedon Jul 3, 2021
p.s. I use conda and
conda install -c conda-forge pandas=1.3.0worked for me.taldcroft commentedon Jul 3, 2021
The problem is actually in
to_pandas(), and I'm surprised that this was previously working. Pandas was noticing that theindexbeing set (an astropyColumn) had anameattribute, even though it presumably knows nothing about astropyColumn.pllim commentedon Jul 4, 2021
Thanks!