Pandas rendering minor ticks in matplotlib sublots with shared axes

[mikepqr]

When the plot method of a dataframe is passed an ax keyword to render to a matplotlib subplot, the minor ticks and their labels are rendered regardless of whether sharex|y=True. The major ticks/labels are correctly removed.

Here's a short notebook that reproduces the issue. Note that the minor ticks are included in the upper axes of the final figure.

I am 90% certain this is a regression, but I wasn't able to pin down which version of pandas or matplotlib prompted this bug. Nor was I able to figure out whether this was due to pandas or matplotlib.

[mikepqr]

To add to this, the bug shown in the notebook linked above happens in

CPython 2.7.10
IPython 3.2.1
pandas 0.16.2
matplotlib 1.4.3

The bug does not happen with:

CPython 2.7.9
IPython 3.2.0
pandas 0.16.0
matplotlib 1.4.3

A pandas regression?

[TomAugspurger]

Thanks. It probably is a maybe a regression on our end. We kind of take over the axes and use our own logic for determining what to draw.

Your plot appears to work when you pass in sharex=True to the .plot calls. Could you try that out.

[mikepqr]

Yes, you're right. sharex=True in the individual .plot calls fixes this.

To be consistent with the matplotlib API this should not probably be necessary.

Happy to attempt a first PR if someone can point me in the direction of the file(s) that might have introduced this regression!

[TomAugspurger]

@sinhrks am I correct in thinking that this is buggy?

In [25]: x = pd.DataFrame(list(zip(range(10), range(0, -10, -1))), columns=['a', 'b'])

In [26]: x.index = pd.date_range('2000-01-01', freq='Q', periods=10)

In [27]: fig, ax = plt.subplots(2, 1, sharex=True)

In [28]: x.plot(subplots=True, ax=ax)
Out[28]:
array([<matplotlib.axes._subplots.AxesSubplot object at 0x116b96630>,
       <matplotlib.axes._subplots.AxesSubplot object at 0x117580198>], dtype=object)

In [29]: plt.savefig('nosavefig.png')

Note that the axis is not shared.

In [30]: fig, ax = plt.subplots(2, 1, sharex=True)

In [31]: x.plot(subplots=True, ax=ax, sharex=True)
/Users/tom.augspurger/Envs/py3/lib/python3.4/site-packages/pandas/tools/plotting.py:3218: UserWarning: When passing multiple axes, sharex and sharey are ignored.These settings must be specified when creating axes
  "These settings must be specified when creating axes", UserWarning)
Out[31]:
array([<matplotlib.axes._subplots.AxesSubplot object at 0x117deb940>,
       <matplotlib.axes._subplots.AxesSubplot object at 0x10c64f828>], dtype=object)

In [32]: plt.savefig('with_share.png')

Now it is shared, but we get the warning telling us to do what we just did.

I haven't looked at this section in a while.

[TomAugspurger]

@WilliamsMJ probably somewhere near here: https://github.com/pydata/pandas/blob/818f0a7e88762e9ac77609024ecb96ea185f4667/pandas/tools/plotting.py#L977

or here: https://github.com/pydata/pandas/blob/818f0a7e88762e9ac77609024ecb96ea185f4667/pandas/tools/plotting.py#L3155

[mikepqr]

I think it's 015fc62 that's the problem.

Given the stated intentions of that commit, it's above my pay grade to determine whether this new behaviour is a feature or a bug!

[sinhrks]

@TomAugspurger Yes, major/minor ticklabel should be displayed when sharex=False (#7801), and hidden otherwise. Unfortunately test cases must have been incomplete.

[TomAugspurger]

cc @JanSchulz

Do you remember in #9740 how the interaction of sharex in plt.subplots and sharex in df.plot is supposed to work?