Can't pickle plots from dataframes with date indexes

[naught101]

Already report this at matplotlib/matplotlib#9828, but it appears that that issue is caused by pandas using a lambda in the format_dateaxis function. I don't know if there is a way around this, but it prevents pandas-produced matplotlib figures from being pickled.

MPL Bug report

Trying to pickle a figure that was produced using pandas.plot(), which includes a datetime index. Not sure if this is a pandas bug, or MPL.

Here's an example:

In [81]: sample
Out[81]: 
time
2002-07-28 08:00:00    0.0
2002-07-28 08:30:00    0.0
2002-07-28 09:00:00    0.0
2002-07-28 09:30:00    0.0
2002-07-28 10:00:00    0.0
2002-07-28 10:30:00    0.0
2002-07-28 11:00:00    0.0
2002-07-28 11:30:00    0.0
2002-07-28 12:00:00    0.0
2002-07-28 12:30:00    0.0
2002-07-28 13:00:00    0.0
2002-07-28 13:30:00    0.0
2002-07-28 14:00:00    0.0
2002-07-28 14:30:00    0.0
2002-07-28 15:00:00    0.0
2002-07-28 15:30:00    0.0
2002-07-28 16:00:00    NaN
2002-07-28 16:30:00    0.0
2002-07-28 17:00:00    0.0
2002-07-28 17:30:00    0.0
Name: Qle_qc, dtype: float64

In [82]: sample.plot()
Out[82]: <matplotlib.axes._subplots.AxesSubplot at 0x7fd5c06a1f60>

In [83]: fig = gcf()

In [84]: pickle.dump(fig, open('test.pkl', 'wb'))
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
/data/documents/uni/phd/projects/FluxnetTrafficLights/scripts/analysis/site_predictability.py in <module>()
----> 1 pickle.dump(fig, open('test.pkl', 'wb'))

AttributeError: Can't pickle local object 'format_dateaxis.<locals>.<lambda>'

In [85]: pickle.dump(sample, open('sample.pkl', 'wb'))

I've tried this with a manually constructed DF, and it worked, so I'm not sure why this is happening. I'm including a pickle of the above sample DF for investigation. It is originally a slice from an xarray, from a netcdf dataset.
sample.pkl.gz

Matplotlib version

Using Conda on linux (kubuntu 17.10):

ipython                   6.1.0                    py36_0  
matplotlib                2.1.0            py36hba5de38_0
pandas                    0.20.3                   py36_0  

MPL backend: Qt5Agg

[Licht-T]

@naught101 Thanks for reporting this issue.
Your data reproduce the error in my environment, but I cannot create the simple code which reproduce this. Do you have any idea?

[Licht-T]

Got it. Simple repro. code. The index which length <= 2 does not repro.

import pandas as pd
from matplotlib import pyplot as plt
import numpy
import pickle
%matplotlib inline

idx = pd.DatetimeIndex(['2002-07-28 08:00:00', '2002-07-28 08:30:00', '2002-07-28 09:00:00'])
s = pd.Series([1, 2, 3], index=idx)

s.plot()
fig = plt.gcf()

pickle.dump(fig, open('test.pkl', 'wb'))

[Licht-T]

Seems that this sets the lambda function.

pandas/pandas/plotting/_timeseries.py

Line 330 in c8dafb5

subplot.format_coord = lambda t, y: (