Subplots and JointGrid

Say I create a figure that will hold multiple plots, one of which I want to be a JointGrid

f = plt.figure(figsize=(12,6))
gs = gridspec.GridSpec(4,1 , height_ratios=[1.5,1.5,1.5, 1.5])
subplots = list(gs)
first_axis = f.add_subplot(subplots[0])

I would like to create a JointGrid plot on this axis. Is there any way to do that?

JointGrid does not seem to accept an ax parameter, I guess because it creates several axes (multiple of them, presumably), i.e. this doesn’t work:

sns.JointGrid(xcol, ycol, df, size=12, ax=first_axis)

How can I use sns.JointGrid with subplot?

[ImportanceOfBeingErnest]

I wrote a small class in this Stackoverflow anser that would allow to transfer a plot from one of the figure type seaborn graphs to a position of a gridspec in a different matplotlib figure. There are some minor issues still, so such solution needs a bit tweaking, but is in principle usable.

The output might then look something like this:

[bendichter]

I have implemented this!

fig = plt.figure(figsize=(16,8))
gs = gridspec.GridSpec(1,2, wspace=.2)

fig.add_subplot(gs[0])
iris = sns.load_dataset("iris")
g = sns.PairGrid(iris, subplot_spec=gs[1])
g = g.map(plt.scatter)

sns.pairplot(iris, grid_kws={'subplot_spec':gs[1]}) also works.

I also implemented it for JointGrid:

fig = plt.figure(figsize=(8,4))
gs = gridspec.GridSpec(1,2, wspace=.2)

fig.add_subplot(gs[0])
g = sns.JointGrid(x="total_bill", y="tip", data=tips, size=5, ratio=2, 
                  subplot_spec=gs[1])
g = g.plot_joint(sns.kdeplot, cmap="Reds_d")
g = g.plot_marginals(sns.kdeplot, color="r", shade=True)

(similar results for g = sns.jointplot(x="total_bill", y="tip", kind='kde',data=tips, size=5, ratio=2, subplot_spec=gs[1]))

The JointPlot/jointplot implementation was pretty straightforward, however, PairGrid/pairplot depends on a change to matplotlib that is in the master branch but has not yet been released. The implementation of plt.subplots() was recently moved to fig.subplots(). That change allowed me to implement this without a giant overhaul to seaborn, because it allowed me to call subplots and use the sharex and sharey optional arguments on a pre-existing figure. To do it with the current release, one would have to implement sharex and sharey themselves. Then we’d probably want to change it to this implementation once the mpl update occurs anyway so I feel like that would be a waste.

I’m new to contributing, so I don’t know what the standard protocol is for features that rely on not-yet-released dependencies. Please advise.