When building pandas 0.19.2 or 0.20.3 I get many similar test failures.
Here is one from the attempt to build 0.19.2:
======================================================================
ERROR: test_timedelta64_equal_timedelta_supported_ops (pandas.tests.series.test_operators.TestSeriesOperators)
----------------------------------------------------------------------
Traceback (most recent call last):
File "/tmp/guix-build-python-pandas-0.19.2.drv-0/pandas-0.19.2/pandas/tests/series/test_operators.py", line 570, in test_timedelta64_equal_timedelta_supported_ops
lhs = op(ser, nptd)
File "/tmp/guix-build-python-pandas-0.19.2.drv-0/pandas-0.19.2/pandas/core/ops.py", line 696, in wrapper
converted = _Op.get_op(left, right, name, na_op)
File "/tmp/guix-build-python-pandas-0.19.2.drv-0/pandas-0.19.2/pandas/core/ops.py", line 330, in get_op
return _TimeOp(left, right, name, na_op)
File "/tmp/guix-build-python-pandas-0.19.2.drv-0/pandas-0.19.2/pandas/core/ops.py", line 343, in __init__
lvalues = self._convert_to_array(left, name=name)
File "/tmp/guix-build-python-pandas-0.19.2.drv-0/pandas-0.19.2/pandas/core/ops.py", line 454, in _convert_to_array
if (inferred_type in ('datetime64', 'datetime', 'date', 'time') or
TypeError: data type "datetime" not understood
I get these errors with python-dateutil 2.6.0 as well as 2.6.1. Could you please confirm that the tests pass when building with python-dateutil 2.6.x?
I'm building with Python 3.5.3.
Comment From: chris-b1
Can you please try on master - we do test with 2.6.1, what you are seeing is most likely fixed by https://github.com/pandas-dev/pandas/pull/16880
Comment From: rekado
Hmm, I applied the patches from #16880, and while that avoids the TypeError, I now get AssertionErrors like this one:
======================================================================
FAIL: pandas.tests.plotting.test_datetimelike.TestTSPlot.test_finder_daily
----------------------------------------------------------------------
Traceback (most recent call last):
File "/gnu/store/341qa9zyviccnwvkw3yb67yligjvizb9-python-nose-1.3.7/lib/python3.5/site-packages/nose/case.py", line 197, in runTest
self.test(*self.arg)
File "/tmp/guix-build-python-pandas-0.20.3.drv-0/pandas-0.20.3/pandas/tests/plotting/test_datetimelike.py", line 409, in test_finder_daily
assert xp == rs
AssertionError
I'm just trying to upgrade the pandas package we provide with GNU Guix, so I'd rather not move to master.
[This is just one of 26 failures (and many more errors) that I'm getting since upgrading to numpy 1.13.x, so maybe this is rather a problem with building with this version of numpy.]
Comment From: chris-b1
0.21.0
will be released soon, so may make sense to test master, it also includes some numpy 1.13 fixes, https://github.com/pandas-dev/pandas/pull/16654 is one, may be some other fixups
Comment From: jreback
pandas is not generally forwards compatible for dependencies. numpy was in the 1.12 IIRC and dateutil was 2.6.0 or less IIRC. (you could actually go back to some of our release builds and check).
cherry-picking specific commits not generally be advisable. You should simply not ship with a higher version of a supported dependency.
Comment From: rekado
@chris-b1 with the latest version from master I still get 19 test failures and 2443 errors.
The errors are things like this:
======================================================================
ERROR: pandas.tests.test_resample.TestResampleAPI.test_getitem
----------------------------------------------------------------------
Traceback (most recent call last):
runTest
self.test(*self.arg)
r = self.frame.resample('H')
AttributeError: 'TestResampleAPI' object has no attribute 'frame'
@jreback Is there an up-to-date list of dependencies and their versions that are known to be supported for a given release?
Comment From: jreback
you would have to look on the CI for the builds that are available at that time; https://travis-ci.org/pandas-dev/pandas/builds
Comment From: rekado
Okay, thank you.
I have since downgraded numpy to 1.12.x.
Comment From: gabegm
Also getting this error on 0.21.0. Downgrading to 0.19.2 fixed this.