Plotting with seaborn using the matplotlib object-oriented interface

Question

I strongly prefer using matplotlib in OOP style:

f, axarr = plt.subplots(2, sharex=True)
axarr[0].plot(...)
axarr[1].plot(...)

This makes it easier to keep track of multiple figures and subplots.

Question: How to use seaborn this way? Or, how to change this example to OOP style? How to tell seaborn plotting functions like lmplot which Figure or Axes it plots to?

Leund

Related: This issue and this answer for a possible workaround, which would allow to create a seaborn figure and move all the axes to a subplot grid later. – Modish

Answer 1

It depends a bit on which seaborn function you are using.

The plotting functions in seaborn are broadly divided into two types:

• "Axes-level" functions, including regplot, boxplot, kdeplot, and many others
• "Figure-level" functions, including relplot, catplot, displot, pairplot, jointplot and one or two others

The first group is identified by taking an explicit ax argument and returning an Axes object. As this suggests, you can use them in an "object oriented" style by passing your Axes to them:

f, (ax1, ax2) = plt.subplots(2)
sns.regplot(x, y, ax=ax1)
sns.kdeplot(x, ax=ax2)

Axes-level functions will only draw onto an Axes and won't otherwise mess with the figure, so they can coexist perfectly happily in an object-oriented matplotlib script.

The second group of functions (Figure-level) are distinguished by the fact that the resulting plot can potentially include several Axes which are always organized in a "meaningful" way. That means that the functions need to have total control over the figure, so it isn't possible to plot, say, an lmplot onto one that already exists. Calling the function always initializes a figure and sets it up for the specific plot it's drawing.

However, once you've called lmplot, it will return an object of the type FacetGrid. This object has some methods for operating on the resulting plot that know a bit about the structure of the plot. It also exposes the underlying figure and array of axes at the FacetGrid.fig and FacetGrid.axes arguments. The jointplot function is very similar, but it uses a JointGrid object. So you can still use these functions in an object-oriented context, but all of your customization has to come after you've called the function.

Madrigal