scatter_plot produces double plots

following #3473:

import pandas as pd
import numpy as np
from pandas.tools.plotting import scatter_plot 

df = pd.DataFrame( np.random.randn(100,4))
scatter_plot(df,x=1,y=2)

**Plot**

**Same Plot Again**

[cpcloud]

What backend are you using? I can't repro this.

just plain ol'

ipython qtconsole --pytlab=inline

[cpcloud]

I see it now. Returning fig is the problem. Returning just the axes would probably be the most consistent with the pandas API.

[cpcloud]

I need some git-fu practice, is it possible for me to fix someone else's PR?

hang on, there's some sort of blood oath with statsmodels in the mists of time on how we do plots...
here it is : #1501 (comment)

no on the latter, unless you're a repo collab on his fork.

[cpcloud]

Ah, ok. Thanks. Well, qtconsole seems to be violating the blood-oath :)

[jseabold]

This "blood oath" was on the advice from the matplotlib and IPython creators on the best and most general way to write library plotting code. Please bring up specific issues in a public forum, so we can continue to coordinate.

That wasn't a criticism, it's just hidden away on an old issue and I wanted to make it known.
Is there anywhere documenting the rationale? would appreciate having somewhere to to point
to when this comes up.

[jseabold]

Didn't mean to sound defensive. Was trying to dispel mysticism. Advice we've been going on.

https://groups.google.com/d/topic/pystatsmodels/biNlCvJPNNY/discussion

added to CONTRIBUTING.md

[jseabold]

FYI, you just need to catch the fig to keep it from displaying (or turning off inline plotting should also work)

fig = plot_func()

[cpcloud]

@jseabold I think only the latter will work, since returning the Figure object is what is causing this bug.

[jseabold]

Both work. Try 'em out. FWIW, I don't consider this a bug.

[cpcloud]

@jseabold I agree, I actually quoted bug but I changed it because I didn't want to sound like a smart-ass.