Accessing handles of plot objects in plot called through pandas

Question

I would like to make use of the stacked bar plot feature of pandas DataFrames. However, I would like to follow up by modifying the bars which were just plotted. For that, I need access to their graphics handles. But all I get back from df.plot() is an axis handle. How do I get a list of what was specifically created in a particular call? This would be necessary, for instance, if there were preexisting objects in the same axes from an earlier plot, and I wanted to differentiate, ie to treat just the new ones. Here is my code which plots twice to the same axis. I would like to end up having the stacked bars clustered next to each other, so I thought I could just make them a little narrower and shift them right and left. But for that I would need handles of what was made in the second call.

import random
import pandas as pd
import numpy as np
from pylab import *
close('all')
df = pd.DataFrame(np.random.random((7, 5)) * 10,               
                        index=list('abcdefg'), columns=list('ABCDE'))

print df

ax = df.plot(kind='bar', stacked=True, align='center')
c1=ax.containers
hh=findobj(ax,lambda xx:hasattr(xx,'width'))
setp(hh,'width',.3)
print hh

df2= pd.DataFrame(np.random.random((6, 5)) * 10,               
                        index=list('abcdef'), columns=list('FGHIJ'))

ax2 = df2.plot(kind='bar', stacked=True, align='center', ax=ax,colormap='bone')
c2=ax2.containers
print ax==ax2 # Should be true
c1==c2 # Should be False
show()

I have tried to get the list of containers ax.containers after the first call, and then again after the second, and taking the set difference, but this didn't work.

Answer 1

You're looking for the patches attribute of axes. You can get the original x positions with [p.get_x() for p in ax.patches]. This gives the left endpoint for each patch.

You'll want to add or subtract some value to each of the patches. Something like: new_xs = [p.get_x() - .5 for p in ax.patches]. And then set_x to the new values:

>>>for p, x in zip(ax.patches, new_xs):
       p.set_x(x)

You'll also probably need to spread the x_ticks out more, or make the width of each patch smaller with p.set_width(). It may just be easier to do it all through matplotlib, rather than trying to adjust what pandas gives back.