Seaborn: Wishlist for sns.scatterplot

[mwaskom]

Open comment thread for functionality that would be good to have in a sns.scatterplot function. Some things that have been mentioned are:

• Scaling point size off a continuous variable that isn't directly interpretable as a point size (cf. #310)
• Better default colors (although #293 would also fix this)
• A better interface for handling continuous point coloring in a FacetGrid context, this currently requires a bit of a hack as color is the fourth positional argument so you would need to pass in a dummy argument for size.
• Jittering x and y points (currently build into regplot)

What else?

[shoyer]

To follow-up on your third bullet (though this is also a bit of separate issue), I've found that continuous point coloring with a FacetGrid/plt.scatter can be _very_ slow when using the hue argument (e.g., it can easily take seconds), because FacetGrid makes separate plots for every color. A faster interface for this would definitely be appreciated.

[mwaskom]

Ah, yeah, currently I wouldn't even say that's something that's supposed to work – everything in FacetGrid is structured so that hue is categorical. I would say that if you want to use a continuous variable you're better off binning it yourself outside of seaborn.

[germannp]

What about some interactivity? Printing the index or the Series upon klicking the point is very useful when you need to check the raw data underlying a certain point. Such functionality is readily available in matplotlib.
This would maybe even be more suitable for sns.pairplot, as this function is particularly handy for quick data exploration.

[BrenBarn]

One thing I often wish for is the ability to specify a label for each point, that would be shown as text next to the point. This is very useful for scatterplots with a relatively small number of points which you want to identify individually (e.g., countries, company names, specific products), and it's currently a pain with matplotlib because you can't pass a vector of strings to a single text call. Ideally the point labels could then be turned on and off with a quick command.

[mwaskom]

It would be possible for a seaborn function to accept a vector of strings and draw them, but it would be bound by the same limitations as when you are using matplotlib directly. So each point would have to be a separate object "behind the scenes" and I don't think it would be possilbe to do what you describe in terms of turning the labels on and off.

[dansondergaard]

In general, it'd be great to be able to change marker size/style depending on the data. E.g.

sns.scatterplot(data, x='PC1', y='PC2', hue='Disease', marker='Sex', size='SampleQuality')

It's currently a drag to do this with sns.lmplot.