Convenience function: DataFrame to linked-brushing plot

[danielballan]

The following is adapts the linked brushing example for a pandas DataFrame, generating NxN axes for DataFrame with N columns. This seems like a common use case, and a useful shortcut for data exploration. Do we want to get into the business of providing convenience functions like this?

def linked_axes(df):
    "Draw an array of axes with linked brushing."
    L = df.shape[1]
    fig, ax = plt.subplots(L, L, sharex="col", sharey="row", figsize=(2*L, 2*L))
    fig.subplots_adjust(left=0.05, right=0.95, bottom=0.05, top=0.95,
                        hspace=0.1, wspace=0.1)

    for i in range(L):
        for j in range(L):
            points = ax[L - 1 - i, j].scatter(df.iloc[:, j], df.iloc[:, i],
                                              c='k', s=40, alpha=0.2)
            if i == 0:
                ax[L - 1 - i, j].set_xlabel(df.columns[j])
            if j == 0:
                ax[L - 1 - i, j].set_ylabel(df.columns[i])

    # remove tick labels
    for axi in ax.flat:
        for axis in [axi.xaxis, axi.yaxis]:
            axis.set_major_formatter(plt.NullFormatter())

    # Here we connect the linked brush plugin
    plugins.connect(fig, plugins.LinkedBrush(points))
    return fig

We could quibble about how the axes labels and ticks are handled, but there won't be a good solution until tick formatting is fully supported.

[aflaxman]

I think that there is great value to providing convenience functions, and the interactivity design makes them reliant enough on the mpld3 package that including them could make sense. OTOH, @jakevdp has had some good instincts about keeping the burden low for package maintenance, and this might cross his threshold for what is in scope.

If we do include it, maybe we should have an mpld3.pandas module for it and things like it. A very different direction would be trying to get this function into pandas, like df.plot(kind='linked_axis').

[jakevdp]

Nice! I think this could be useful. I do have some qualms about extending the scope of mpld3 like this, though, as @aflaxman mentioned...

Maybe we could compromise and create a separate repository: something like mpld3_addons that could live in the [mpld3])(http://github.com/mpld3) github group and be pip-installable. The advantage there is that the release cycle wouldn't be tied to mpld3 (which is currently sluggish and will likely get more-so!) so that these sorts of tools could be developed and released more quickly.

[danielballan]

Yes, this might make more sense in pandas, but I expect pandas would wait for mpld3 to be more stable and complete before adopting it as an optional dependency.

I agree that admitting recipes like this one as mpld3 functions might lead to feature bloat. How about mpld3_cookbook? If we ultimately decide not to maintain a separate repo, something like the pandas cookbook might be worthwhile -- example designed for general reusability over pedagogical value.

[jakevdp]

Sounds great @danielballan! I think you have admin rights to the mpld3 group - would you like to start on the mpld3_cookbook repo?

[danielballan]

Sure thing!