can jointplot visualize different classes?

[hdmetor]

Is it possible to have different color for different classes in joinplot? Something like

df = pd.DataFrame(np.random.randn(100, 4), columns=list('ABCD'))
df['category'] = np.random.randint(0, high  = 5, size = 100)
sns.jointplot('B','A', data = df, color='category')

where a different color is used for each different value in the category column. Even better, would be possible to have a color for each value after a function is mapped to the column?
In my example, assume I would like to have a color for odd values, and another one for even values of the column, i.e.

sns.jointplot('B','A', data = df, color='category', map = (lambda x: x % 2))

[mwaskom]

Not easily, no.

[mwaskom]

Actually I guess that's not strictly true, for kind=scatter you can pass a vector of colors to c in scatter_kws. But it doesn't have general hue-based grouping like lmplot does.

[mwaskom]

See my answer to #294 for more ideas on what you could do.

[hdmetor]

Ok, thanks a lot.
I would be a cool feature to add

[mwaskom]

It could be useful, although for several of the jointplot styles (like hex or kde) it wouldn't make a lot of sense. JointGrid is pretty straightforward to use directly so I don't want to add a lot of complexity to jointplot right now.

[ruxi]

I wrote a gist on how to do this:

https://gist.github.com/ruxi/ff0e9255d74a3c187667627214e1f5fa

[efharkin]

@mwaskom would you consider a PR to add support for hue in jointplot and/or JointGrid, or at least some changes to JointGrid.plot_{joint,marginals} to support some of the categorical plotters? (As of now, joint_grid.plot_marginals(sns.boxplot, hue=whatever) raises an error about an unexpected keyword argument used to set the orientation. Seems like an easy fix and I'd be happy to work on this.)

It seems like there's a lot of demand for increased support for visualizing the distributions of categorical data (#861, #1486, this question). @ruxi's gist is a nice solution, but it would be great to have this behaviour or something similar directly in jointplot/JointGrid. A potential drawback of folding this behaviour directly into JointGrid is that the joint axis might not look very good if we use certain styles (eg hex), but I think this could be solved by raising an error.

Let me know if this fits with how you'd like to see jointplot and JointGrid used and I can work on a PR.

[mwaskom]

xref #2210