CI: mpl test failures

[jreback]

https://travis-ci.org/pandas-dev/pandas/jobs/463295292; keep failing even with a restart.

=================================== FAILURES ===================================
_____________________ TestTSPlot.test_is_error_nozeroindex _____________________
[gw0] linux -- Python 3.6.6 /home/travis/miniconda3/envs/pandas-dev/bin/python
self = <pandas.tests.plotting.test_datetimelike.TestTSPlot object at 0x7f049594ef98>
    def test_is_error_nozeroindex(self):
        # GH11858
        i = np.array([1, 2, 3])
        a = DataFrame(i, index=i)
>       _check_plot_works(a.plot, xerr=a)
pandas/tests/plotting/test_datetimelike.py:89: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
pandas/tests/plotting/test_datetimelike.py:1562: in _check_plot_works
    pickle.dump(fig, path)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
m = <bound method AffineBase.transform of <matplotlib.transforms.BlendedAffine2D object at 0x7f0495e0e390>>
    def _pickle_method(m):
        """Handle pickling issues with class instance."""
>       if m.im_self is None:
E       AttributeError: 'function' object has no attribute 'im_self'
../../../miniconda3/envs/pandas-dev/lib/python3.6/site-packages/statsmodels/graphics/functional.py:32: AttributeError
________________________ TestTSPlot.test_timedelta_plot ________________________
[gw0] linux -- Python 3.6.6 /home/travis/miniconda3/envs/pandas-dev/bin/python
self = <pandas.tests.plotting.test_datetimelike.TestTSPlot object at 0x7f04951ab208>
    def test_timedelta_plot(self):
        # test issue #8711
        s = Series(range(5), timedelta_range('1day', periods=5))
        _, ax = self.plt.subplots()
>       _check_plot_works(s.plot, ax=ax)
pandas/tests/plotting/test_datetimelike.py:1450: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
pandas/tests/plotting/test_datetimelike.py:1562: in _check_plot_works
    pickle.dump(fig, path)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
m = <bound method AffineBase.transform of <matplotlib.transforms.BlendedAffine2D object at 0x7f0494f0b160>>
    def _pickle_method(m):
        """Handle pickling issues with class instance."""
>       if m.im_self is None:
E       AttributeError: 'function' object has no attribute 'im_self'
../../../miniconda3/envs/pandas-dev/lib/python3.6/site-packages/statsmodels/graphics/functional.py:32: AttributeError

[jreback]

cc @TomAugspurger

[TomAugspurger]

Thanks... taking a look now.

[TomAugspurger]

@tacaswell what kinds of guarantees does matplotlib make for pickling figures?

[h-vetinari]

It might be related to this statsmodel bug: statsmodels/statsmodels#4772
Unfortunately, it doesn't look like a statsmodel 0.10 is on the horizon: https://github.com/statsmodels/statsmodels/milestone/11

The last comment in that thread seems to provide a workaround.

[TomAugspurger]

Weird... thanks @h-vetinari. I'd rather have to carry that patch around. How about skipping in that case?

        if PY3 and not 'statsmodels' in sys.modules:
            with ensure_clean(return_filelike=True) as path:
                pickle.dump(fig, path)