ENH: Double plotting issue

[aflaxman]

This one-line change addresses the issue of two copies of the traceplot appearing when using pm.traceplot(trace) in the ipython notebook, as described in #243. I know there is a simple workaround described there, but I would prefer this to "just work". I couldn't find any code where the returned fig was used, and if you have some, you can use plt.gcf() to get a reference to the current figure after calling pm.traceplot.

[jsalvatier]

Oh yeah, that seems much better! Didn't know about that. The double plotting issue was annoying.

However, can you rebase your change on top of master instead of merging it?

[jsalvatier]

You could also do

git checkout master
git checkout -b doubleplot
git cherry-pick c6ad789

[fonnesbeck]

I'm not sure I'm on board with this one. A limitation with the PyMC 2 plots is that they are essentially unmodifiable because the functions do not return the figure object when plotted. The nice thing about the plots in PyMC 3 is that with the figure object returned, you can add elements to the plot that are non-standard. It may not sound like a big deal, but when it comes time to use figures for publications, etc., one often requires some customization. All you have to do to prevent double-plotting with the current implementation is to either assign the output to a variable, or use a semi-colon at the end of the call.

One alternative would be to allow for an axis argument on the plots to which one could pass a set of axes that could subsequently be modified.

[jsalvatier]

hmm, good point. We could also have an argument that causes the value to be returned. Or we could always return it in a tuple or something so it doesn't get plotted the second time.

[jsalvatier]

How do the plotting libraries work here? with %pylab inline plot seems to return a value but not get double plotted.

[fonnesbeck]

Yeah, returning (fig, ax) should do the trick.

[fonnesbeck]

See if the fix in double_plot_fix fits the bill.

[jsalvatier]

The reason why out plots double plot and pylab plots dont is that we're returning a matplotlib.figure.Figure and they're returning objects like [matplotlib.lines.Line2D at 0x104b492c].

[jsalvatier]

@fonnesbeck the double plot fix looks good to me, though you're overwriting some other changes.

[aflaxman]

@fonnesbeck your double_plot_fix works fine for me. Happy to have that instead, if you think the returned stuff will be of use.

[fonnesbeck]

@jsalvatier that's bizarre. Not sure how that happened. Will fix.

[fonnesbeck]

Closing this in favor of #745