BUG: Boxplot to return axes, not dict

[phobson]

Fixes #4264.

[jreback]

• can you hook up travis
• test
• release notes entry

is this consistent with #4020 (I believe that's the point here!)

[phobson]

@jreback

• Hooked up Travis
• I don't think there's much testing to add here. I'm not sure we even want to keep the last commit i just pushed. I think it has much larger implications about all of the plotting functions that we really don't want to deal with in the simple PR that deals with just a simple issue
• Release notes coming after we decide on the direction to take re: my last "TST" commit

[cpcloud]

@phobson This should be kept as plt.Artist since not all plotting functions return Axes instances.

[cpcloud]

@phobson Can you remove the dict check here, since boxplot was the only function returning a dict?

[jreback]

@cpcloud can u take a look

[cpcloud]

Looks okay to me, modulo what I said in the outdated diff.

@jreback Do you think it's worth noting in the docs how to recreate the dict that's returned as of now? matplotlib returns a dict as well so users coming from there might expect a dict.

[phobson]

@cpcloud
The more I think about this, the more I think that boxplot should return the ax to be consistent with the other functions, but there is value in the dict of Artists as well. Namely, boxplots of out of the -- uh-- box are pretty ugly. I always keep a function like this handy:

def formatBoxplot(bp, color='b', marker='o', linestyle='-'):
    # format the box itself
    for box in bp['boxes']:
        plt.setp(box, color='k', linewidth=0.75, zorder=4)

    # format the medians
    for med in bp['medians']:
        plt.setp(med, linewidth=1.00, color=color, linestyle=linestyle, zorder=2)

    # format the whiskers
    for whi in bp['whiskers']:
        plt.setp(whi, linestyle='-', color='k', linewidth=0.75, zorder=4)

    # format the caps on top of the whiskers
    for cap in bp['caps']:
        plt.setp(cap, linewidth=0.75, zorder=4)

    # format the outliers
    for fli in bp['fliers']:
        plt.setp(fli, marker=marker, markersize=4, zorder=4,
                 markerfacecolor='none', markeredgecolor=color)

Any thoughts on returning both the Axes and the dict.

[cpcloud]

@phobson I was thinking something similar after glancing at the matplotlib source, sigh. I'm not sure there's a point to return both the axes and the dict since you can get the axes easily with gca() or get it from one of the lines in the dict (although this annoying so maybe return both might be better here)

[cpcloud]

now that i think about it, i think both is an okay compromise

[phobson]

Since Travis is failing and I'm not at a machine where I can easily build this, I'll have to pick this up again later.

[jreback]

@phobson ping!

[phobson]

Get folks. Sorry for the delay. When started poking around the source and tests, this was a lite more complicated than I anticipated. I hope to get to this soon. (Maybe tonight, or by the end of the week.)

Paul Hobson
Sorry if this is unintelligible. I'm on my phone.

On Wed, Sep 25, 2013 at 4:59 PM, jreback notifications@github.com wrote:

@phobson ping!

Reply to this email directly or view it on GitHub:
#4472 (comment)

[seclinch]

The documentation seems to suggest that pandas.DataFrame.boxplot already returns a matplotlib.axes.AxesSubplot object -- but I don't believe this to be the case?

Perhaps the documentation could be updated to indicate that boxplot() currently returns a dict?

[jtratner]

@seclinch - it'd be great if you could submit a pull request with the docs change. (which we'll revert once this current pull request is merged)

[jreback]

@cpcloud is this correct?

@phobson maybe a test to validate that the box plot is in fact returning a type of Axis (and not something else?)

[jreback]

@phobson can you rebase this

[jreback]

@phobson can you rebase.... @cpcloud pls review

[jreback]

@TomAugspurger @cpcloud

what do you think?

[cpcloud]

@phobson can you add a doc saying how to get the dict back? Then I think good to go.

[jreback]

does this fix #4636 as well?

[phobson]

I don't think this is ready yet. I'll have some bandwidth on friday to look
into it though. Last I remember I was having trouble groking the tests.

On Wed, Apr 9, 2014 at 5:56 AM, jreback notifications@github.com wrote:

does this fix #4636 #4636 as
well?

Reply to this email directly or view it on GitHubhttps://github.com//pull/4472#issuecomment-39960179
.

[cpcloud]

Let us know if we can help.

[cpcloud]

There's a hacky plot testing function that special cases checking the return value type of boxplot. You'll have to remove the lines that check for a dict and then the tests should pass.

[jreback]

@phobson status on this?

@TomAugspurger @cpcloud

[jreback]

@phobson ok....going to do this for 0.14.

[TomAugspurger]

See @jseabold's comment here #4264 (comment)

I agree that box plot should always return an axes (array of axes if subplots). An optional kwarg return_dict=False (default) to return a dictionary in addition to the axis will allow for more formatting.

[jreback]

superseded by #7096