The seaborn.objects Interface

Author

David Gerard

Published

June 3, 2025

Learning Objectives

Data visualization with seaborn.objects.
https://seaborn.pydata.org/tutorial/objects_interface.html

Python Overview

In R I Want	In Python I Use
Base R	numpy
dplyr/tidyr	pandas
ggplot2	matplotlib/seaborn

I previously taught how to use the basic seaborn interface. But in 2022 the author introduced an interface more similar to ggplot which seems to be the future of the package. So here is a lecture on the seaborns.objects interface.
This interface is experimental, doesn’t have all of the features you would need for data analysis, and will likely change. But I think it looks cool.

Import Matplotlib and Seaborn, and Load Dataset

library(ggplot2)
data("mpg")

All other code will be Python unless otherwise marked.

import matplotlib.pyplot as plt # base plotting functionality
import seaborn as sns           # Original interface
import seaborn.objects as so    # ggplot2-like interface
mpg = r.mpg

Basics

Use so.Plot() to instantiate a Plot object and define asthetic mappings.
- Just like ggplot().
Use the .add() method of the Plot object to add geometric objects and a statistical transformation.
As in ggplot, each aesthetic mapping is followed by a statistical transformation before plotting. But unlike in ggplot, you need to specify this statistical transformation manually.
- E.g. aggregating a categorical variable into a counts before plotting those counts on the y-axis.
- E.g. binning a quantitative variable to make a histogram.
You specify the statistical transformation as the second argument in .add().

One Quantitative Variable: Histogram

Notice that we need to specify the so.Hist() statistical transformation to generate a histogram.

We use the so.Bars() geometric object to plot it after the statistical transformation.

pl = (
  so.Plot(mpg, x = "hwy")
  .add(so.Bars(), so.Hist(bins = 10))
)
pl.show()

One Categorical Variable: Barplot

We use the so.Bar() geometric object after the so.Count() statistical transformation.

pl = (
  so.Plot(mpg, x = "class")
  .add(so.Bar(), so.Count())
)
pl.show()

so.Bar() (for categorical data) and so.Bars() (for quantitative data) seem to be only slightly different based on the defaults.

Dodging Barplots

If you are creating two barplots, annotated by color, you need to be explicit that the bars should dodge eachother with a so.Dodge() transformation.
```
pl = (
  so.Plot(mpg, x = "class", color = "drv")
  .add(so.Bar(), so.Count(), so.Dodge())
)
pl.show()
```

One Quantitative Variable, One Categorical Variable: Boxplot

This interface is currently (November 2022) missing boxplotting functions, so you need to use the old interface.
```
plt.clf()
sns.boxplot(data = mpg, x = "class", y = "hwy")
plt.show()
```
```
plt.clf()
```

I think this is the closest thing to a boxplot you can get right now:

pl = (
  so.Plot(mpg, x = "class", y = "hwy")
  .add(so.Dash(width = 0.4), so.Perc())
  .add(so.Range())
  .add(so.Range(), so.Perc([25, 75]), so.Shift(x=0.2))
  .add(so.Range(), so.Perc([25, 75]), so.Shift(x=-0.2))
)
pl.show()

Two Quantitative Variables: Scatterplot

Base scatterplot uses the so.Dots() geometric object:

pl = (
  so.Plot(mpg, x = "displ", y = "hwy")
  .add(so.Dots())
)
pl.show()

Use the so.Jitter() statistical transformation to make a jittered scatterplot.

pl = (
  so.Plot(mpg, x = "displ", y = "hwy")
  .add(so.Dots(), so.Jitter(1))
)
pl.show()

Use so.Line() (geometric object) and so.PolyFit() (statistical transformation) to add a smoother.

pl = (
  so.Plot(mpg, x = "displ", y = "hwy")
  .add(so.Dots())
  .add(so.Line(), so.PolyFit())
)
pl.show()

I don’t think it does lowess or gam or splines yet, but just a polynomial, which is not optimal. You can control the order of the polynomial by the order argument.

pl = (
  so.Plot(mpg, x = "displ", y = "hwy")
  .add(so.Dots())
  .add(so.Line(), so.PolyFit(order = 1))
)
pl.show()

Annotate by a third variable by adding a color mapping:

pl = (
  so.Plot(mpg, x = "displ", y = "hwy", color = "drv")
  .add(so.Dots())
  .add(so.Line(), so.PolyFit(order = 1))
)
pl.show()

Faceting

Facet by the .facet() method.

pl = (
  so.Plot(mpg, x = "displ", y = "hwy")
  .facet(row = "drv")
  .add(so.Dots())
)
pl.show()

Customizing Look

You can change the scaling using the .scale() method. E.g. here is a \(\log_2\) transformation for the \(x\)-axis.

pl = (
  so.Plot(mpg, x = "displ", y = "hwy")
  .add(so.Dots())
  .add(so.Line(), so.PolyFit(order = 1))
  .scale(x = "log2")
)
pl.show()

You can change the labels by .label().

pl = (
  so.Plot(mpg, x = "displ", y = "hwy")
  .add(so.Dots())
  .label(x = "Displacement (L)", y = "Highway MPG")
)
pl.show()

You can change the theme using .theme(). But it is a little verbose right now.

from seaborn import axes_style
pl = (
  so.Plot(mpg, x = "displ", y = "hwy")
  .add(so.Dots())
  .add(so.Line(), so.PolyFit(order = 1))
  .theme({**axes_style("whitegrid"), "grid.linestyle": ":"})
)
pl.show()

Exercises

Consider the palmer penguins data, which you can load via

penguins = sns.load_dataset("penguins")
penguins.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 344 entries, 0 to 343
Data columns (total 7 columns):
 #   Column             Non-Null Count  Dtype  
---  ------             --------------  -----  
 0   species            344 non-null    object 
 1   island             344 non-null    object 
 2   bill_length_mm     342 non-null    float64
 3   bill_depth_mm      342 non-null    float64
 4   flipper_length_mm  342 non-null    float64
 5   body_mass_g        342 non-null    float64
 6   sex                333 non-null    object 
dtypes: float64(4), object(3)
memory usage: 18.9+ KB

Make a visualization of bill length versus bill depth, annotated by species.
Add OLS lines to for each species to the same plot object you created in part 1 (don’t rerun so.Plot()).
Use pandas.cut() to convert body mass into five equally spaced levels.
Facet your plot from part 2 by the above transformation. You will have to redo the object since we are using a different data frame here.
Make a visualization for the number of each species in the dataset. Make sure you have good labels.