[python] Saving plots (AxesSubPlot) generated from python pandas with matplotlib's savefig

I'm using pandas to generate a plot from a dataframe, which I would like to save to a file:

dtf = pd.DataFrame.from_records(d,columns=h)
fig = plt.figure()
ax = dtf2.plot()
ax = fig.add_subplot(ax)
fig.savefig('~/Documents/output.png')

It seems like the last line, using matplotlib's savefig, should do the trick. But that code produces the following error:

Traceback (most recent call last):
  File "./testgraph.py", line 76, in <module>
    ax = fig.add_subplot(ax)
  File "/opt/local/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/matplotlib/figure.py", line 890, in add_subplot
    assert(a.get_figure() is self)
AssertionError

Alternatively, trying to call savefig directly on the plot also errors out:

dtf2.plot().savefig('~/Documents/output.png')


  File "./testgraph.py", line 79, in <module>
    dtf2.plot().savefig('~/Documents/output.png')
AttributeError: 'AxesSubplot' object has no attribute 'savefig'

I think I need to somehow add the subplot returned by plot() to a figure in order to use savefig. I also wonder if perhaps this has to do with the magic behind the AxesSubPlot class.

EDIT:

the following works (raising no error), but leaves me with a blank page image....

fig = plt.figure()
dtf2.plot()
fig.savefig('output.png')

EDIT 2: The below code works fine as well

dtf2.plot().get_figure().savefig('output.png')

The gcf method is depricated in V 0.14, The below code works for me:

plot = dtf.plot()
fig = plot.get_figure()
fig.savefig("output.png")

It seems easy for me that use plt.savefig() function after plot() function:

import matplotlib.pyplot as plt
dtf = pd.DataFrame.from_records(d,columns=h)
dtf.plot()
plt.savefig('~/Documents/output.png')

You can use ax.figure.savefig(), as suggested in a comment on the question:

import pandas as pd

df = pd.DataFrame([0, 1])
ax = df.plot.line()
ax.figure.savefig('demo-file.pdf')

This has no practical benefit over ax.get_figure().savefig() as suggested in other answers, so you can pick the option you find the most aesthetically pleasing. In fact, get_figure() simply returns self.figure:

# Source from snippet linked above
def get_figure(self):
    """Return the `.Figure` instance the artist belongs to."""
    return self.figure

this may be a simpler approach:

(DesiredFigure).get_figure().savefig('figure_name.png')

i.e.

dfcorr.hist(bins=50).get_figure().savefig('correlation_histogram.png')

So I'm not entirely sure why this works, but it saves an image with my plot:

dtf = pd.DataFrame.from_records(d,columns=h)
dtf2.plot()
fig = plt.gcf()
fig.savefig('output.png')

I'm guessing that the last snippet from my original post saved blank because the figure was never getting the axes generated by pandas. With the above code, the figure object is returned from some magic global state by the gcf() call (get current figure), which automagically bakes in axes plotted in the line above.