Pandas hexbin plot lacks X axis labels in IPython Notebook

[levitsky]

When using pandas.DataFrame.plot() with kind='hexbin', the resulting plot sometimes has no X tick labels and on X axis label. This is observed in IPython Notebook (ipython version 3.2.0) with Python 3.4, matplotlib 1.4.3, with %pylab --no-import-all inline.
The same code produces a correct plot in IPython terminal on the same setup, with %pylab --no-import-all. This happens only on some dataframes, here's the code I run:

df = pd.read_csv('/tmp/example.csv', index_col=0)
df.plot('x', 'y', kind='hexbin', gridsize=10)

Here's the df:

           x           y
0  18.378351  111.080217
1  17.371567  118.277121
2  24.659952  169.528504
3  31.774163  310.880044
4  17.181311  113.618851
5  20.455473  149.341431
6  18.218108  132.492567
7  23.453792  157.083139
8  25.340805  158.190820

And here's the example.csv file used. The plot from the Notebook is below. Also, here is another fellow with the same problem.

[takluyver]

Failed to reproduce on IPython master, Python 3.4, mpl 1.4.2 and pandas 0.15.0.

[levitsky]

I have reproduced it with IPython master (is that correct that I need jupyter/notebook to run the notebook server?).
I installed both from git master into a virtualenv (happened to be Python 2.7) and still see the same problem. Here's IPython.sys_info() from the notebook:

{'commit_hash': u'3e49b3c',
 'commit_source': 'installation',
 'default_encoding': 'UTF-8',
 'ipython_path': '/home/lev/test_venv/lib/python2.7/site-packages/IPython',
 'ipython_version': '4.0.0-dev',
 'os_name': 'posix',
 'platform': 'Linux-4.0.7-2-ARCH-x86_64-with-glibc2.2.5',
 'sys_executable': '/home/lev/test_venv/bin/python2',
 'sys_platform': 'linux2',
 'sys_version': '2.7.10 (default, May 26 2015, 04:16:29) \n[GCC 5.1.0]'}

And here's the whole notebook.

[takluyver]

What's the pandas version? Could this be something that has changed in pandas?

[levitsky]

Pandas version is 0.16.2. It might be dependent on pandas version for all I know (which is next to nothing), all I can say is the plot looks fine when plotted from IPython console, so it must be notebook-related and it manifests with 3.2 and 4.0-dev, at least for me.

[takluyver]

It must be notebook related, but it might be something in pandas or mpl that is just triggered by the notebook.

@jreback, @tacaswell, any idea what might cause the X-axis labels and tick labels to go missing like this?

[levitsky]

Another thing: the labels don't disappear with colobar=False. Updated the linked gist/notebook.

[tacaswell]

That is odd...

x, y = np.random.randn(2, 122)
fig, ax = plt.subplots()
hx = ax.hexbin(x, y, cmap='Greens', gridsize=10)
fig.colorbar(hx)
plt.show()

works correctly in the notebook with both inline and notebook.

[levitsky]

Yes, it only happens when called from DataFrame.plot().
However, when doing ax = df.plot('x', 'y', kind='hexbin', gridsize=10), ax.get_xlabel() returns a non-empty value as expected, and ax.get_xticklabels() also returns a correct list, although labels are not displayed.

[BDannowitz]

I'm seeing this problem as well with python 2.7, pandas 0.16.2.dev, and ipython version 3.1.0.

Looking to the documentation for pandas.DataFrame.plot, the only reference to the x labels and ticks being invisible is in the sharex option. So, I manually set sharex=False (which it should have been by default), and poof, the xlabel and xticks appeared.

Hopefully this helps tracking down the bug.

[takluyver]

Thanks @BDannowitz . That makes it sound like there's an issue in pandas - does someone want to file an issue there? You can reference this issue for background.

[takluyver]

Closing as it appears to be a pandas bug. Anyone who's still seeing it, follow along with pandas-dev/pandas#10678.