Visualization with `plotnine` in `Python`

Visualization is a key part of statistical analyses, especially in systems engineering. In this tutorial, we'll learn to visualize data with plotnine (the Python port of ggplot2).

Please follow along using the code below!

Getting Started

Loading Packages

import pandas as p
from plotnine import *
from gapminder import gapminder as gapminder

# diamonds bundled with plotnine
from plotnine.data import diamonds

Gapminder data

# View it in the console
gapminder

# Glimpse-like summary
gapminder.dtypes, gapminder.shape

Your first scatterplot

# make the folder we save figures into
import os; os.makedirs("plotnine_figures", exist_ok=True)
p1 = ggplot(data=gapminder, mapping=aes(x='gdpPercap', y='lifeExp'))
p1.save("plotnine_figures/01_scatter_base.png", dpi=200, width=6, height=4)

Blank scatterplot base layer with GDP per capita mapped to x and life expectancy mapped to y, before any points are drawn

Add points with + geom_point().

p2 = (ggplot(gapminder, aes(x='gdpPercap', y='lifeExp')) + geom_point())
p2.save("plotnine_figures/01_scatter_points.png", dpi=200, width=6, height=4)

Scatterplot of GDP per capita versus life expectancy with points plotted for every country-year

LC 01

What kind of relationship does this graph show? Why might it matter to policymakers?

Look at the shape of the point cloud as GDP per capita rises: does life expectancy keep climbing at the same rate the whole way across, or does the curve change shape partway through?

As wealth per person (GDP per capita) increases, life expectancy rises quickly then tapers off. This shows a strong relationship between wealth and health.

Transparency (alpha)

p3 = (ggplot(gapminder, aes(x='gdpPercap', y='lifeExp')) + geom_point(alpha=0.2))
p3.save("plotnine_figures/01_scatter_alpha02.png", dpi=200, width=6, height=4)
p4 = (ggplot(gapminder, aes(x='gdpPercap', y='lifeExp')) + geom_point(alpha=0.5))
p4.save("plotnine_figures/01_scatter_alpha05.png", dpi=200, width=6, height=4)
p5 = (ggplot(gapminder, aes(x='gdpPercap', y='lifeExp')) + geom_point(alpha=1))
p5.save("plotnine_figures/01_scatter_alpha1.png", dpi=200, width=6, height=4)

Scatterplot of GDP per capita versus life expectancy with point transparency alpha set to 0.5

Scatterplot of GDP per capita versus life expectancy with point transparency alpha set to 1 (fully opaque)

LC 02

What happens when you change alpha across the three visuals above?

alpha is a geom_point() argument that controls point opacity. Compare what happens to overlapping points as the value moves from near 0 toward 1.

alpha controls transparency from 0 to 1. Higher values are more opaque; lower values are more transparent.

Color: constant vs mapped

# Single color
p6 = (ggplot(gapminder, aes(x='gdpPercap', y='lifeExp')) +
  geom_point(alpha=0.5, color='steelblue'))
p6.save("plotnine_figures/01_scatter_single_color.png", dpi=200, width=6, height=4)

# Color mapped by continent
p7 = (ggplot(gapminder, aes(x='gdpPercap', y='lifeExp', color='continent')) +
  geom_point(alpha=0.5))
p7.save("plotnine_figures/01_scatter_mapped_color.png", dpi=200, width=6, height=4)

Scatterplot of GDP per capita versus life expectancy with all points set to a single steelblue color

Scatterplot of GDP per capita versus life expectancy with point color mapped to continent

LC 03

Where do you place color for a single color vs. multiple colors based on a variable?

Ask whether the color is a fixed property applied to every point (a single value) or is being determined by a column in the data (a mapped aesthetic) — that decides whether it belongs inside aes() or outside it.

Single color: set color inside geom_point(color='...') (outside aes). Mapped colors: set color inside aes(color='variable').

Improving our visualizations

p8 = (ggplot(gapminder, aes(x='gdpPercap', y='lifeExp', color='continent')) +
  geom_point(alpha=0.5) +
  labs(x='GDP per capita (USD)',
       y='Life Expectancy (years)',
       color='Continent',
       title='Does Wealth affect Health?',
       subtitle='Global Health Trends by Continent',
       caption='Points display individual country-year observations.'))
p8.save("plotnine_figures/01_scatter_improved.png", dpi=200, width=8, height=6)

Scatterplot of GDP per capita versus life expectancy colored by continent with title, subtitle, axis labels, and caption added

You can save visuals as objects to reuse them.

myviz = (ggplot(gapminder, aes(x='gdpPercap', y='lifeExp', color='continent')) +
  geom_point(alpha=0.5) +
  labs(x='GDP per capita (USD)', y='Life Expectancy (years)', color='Continent',
       title='Does Wealth affect Health?', subtitle='Global Health Trends by Continent',
       caption='Points display individual country-year observations.'))

myviz.save("plotnine_figures/01_myviz_default.png", dpi=200, width=8, height=6)
(myviz + theme_bw()).save("plotnine_figures/01_myviz_bw.png", dpi=200, width=8, height=6)
(myviz + theme_dark()).save("plotnine_figures/01_myviz_dark.png", dpi=200, width=8, height=6)
(myviz + theme_classic()).save("plotnine_figures/01_myviz_classic.png", dpi=200, width=8, height=6)

The saved myviz scatterplot rendered with plotnine's default theme

The saved myviz scatterplot rendered with theme_bw()

The saved myviz scatterplot rendered with theme_dark()

The saved myviz scatterplot rendered with theme_classic()

LC 04

Recalling a saved plot

Now run myviz on its own - what happens?

myviz

myviz stores the entire ggplot object, not just a picture - so you can display it again, or add more layers onto it with +, without rebuilding it.

When you save a plotnine plot to an object, eg. naming it myviz, you can call up the visual again as many times as you want by just running the myviz object, just like any other object. In a notebook, the object displays itself; in a script, use print(myviz) or myviz.save(...). Because it is still a live plot object, myviz + theme_bw() returns a new plot with that theme, leaving myviz untouched.

Visualizing `diamonds` data

diamonds.head(3)

diamonds.dtypes, diamonds.shape

Boxplots by cut

p9 = (ggplot(diamonds, aes(x='cut', y='price', group='cut')) + geom_boxplot())
p9.save("plotnine_figures/01_boxplot_default.png", dpi=200, width=6, height=4)

Boxplot of diamond price by cut with plotnine's default boxplot fill

p10 = (ggplot(diamonds, aes(x='cut', y='price', group='cut')) +
  geom_boxplot(fill='steelblue'))
p10.save("plotnine_figures/01_boxplot_fill.png", dpi=200, width=6, height=4)

Boxplot of diamond price by cut with every box filled a single steelblue color

p11 = (ggplot(diamonds, aes(x='cut', y='price', group='cut', fill='cut')) +
  geom_boxplot())
p11.save("plotnine_figures/01_boxplot_mapped_fill.png", dpi=200, width=6, height=4)

Boxplot of diamond price by cut with fill color mapped to the cut variable

LC 05

Why do the two boxplot versions look different? What changed in the code to create those effects?

Compare where fill='steelblue' sits in the geom_boxplot() call versus where fill='cut' sits — one is a fixed argument to the geom, the other lives inside aes().

Constant fill uses geom_boxplot(fill='steelblue'). Mapped fill uses aes(fill='cut') to color by variable.

Visualizing Distributions

p12 = (ggplot(diamonds, aes(x='price', fill='cut')) +
  geom_histogram(color='white') +
  labs(x='Price (USD)', y='Frequency', title='US Diamond Sales'))
p12.save("plotnine_figures/01_histogram_default.png", dpi=200, width=8, height=6)

Stacked histogram of diamond price filled by cut, default binwidth

Try adjusting binwidth and theme

p13 = (ggplot(diamonds, aes(x='price', fill='cut')) +
  geom_histogram(color='white', binwidth=500) +
  theme_classic() +
  labs(x='Price (USD)', y='Frequency', title='US Diamond Sales (binwidth=500)'))
p13.save("plotnine_figures/01_histogram_binwidth.png", dpi=200, width=8, height=6)

Stacked histogram of diamond price filled by cut with binwidth set to 500 and theme_classic() applied

LC 06

Make a histogram of price with a narrower binwidth and apply a different theme. Which choices improve readability?

binwidth in geom_histogram() controls how wide each bar's price range is — smaller values show more detail. Try swapping in theme_bw(), theme_dark(), or theme_classic() to change the background and gridlines.

p14 = (ggplot(diamonds, aes(x='price', fill='cut')) +
  geom_histogram(color='white', binwidth=250) +
  theme_bw() +
  labs(x='Price (USD)', y='Frequency'))
p14.save("plotnine_figures/01_histogram_answer.png", dpi=200, width=8, height=6)

Stacked histogram of diamond price filled by cut with binwidth set to 250 and theme_bw() applied

LC 07

Shape of the diamond price distribution

Are most diamonds cheap or expensive? What type of distribution would you call this?

A Normal?

B Uniform?

C Left Skewed?

D Right Skewed?

Picture the diamonds histogram: is the tall clump of bars on the low-price end or the high-price end, and which side has the long thin tail? Comparing diamonds['price'].median() to diamonds['price'].mean() is a quick numeric check.

This is a strongly right-skewed distribution, because the majority of the distribution leans to the left (the clump of the data), while it has a long tail that skews to the right. The median is typically less than the mean in a right skewed distribution - and sure enough, diamonds['price'].median() comes in well below diamonds['price'].mean().

Conclusion

You learned how to build scatterplots, boxplots, and histograms with plotnine and how to control transparency, color mapping, labels, and themes.

Visualization with plotnine in Python