Data Visualization

In this lesson, we'll learn two data visualization libraries matplotlib and seaborn. By the end of this lesson, students will be able to:

• Skim library documentation to identify relevant examples and usage information.
• Apply seaborn and matplotlib to create and customize relational and regression plots.
• Describe data visualization principles as they relate the effectiveness of a plot.

Just like how we like to import pandas as pd, we'll import matplotlib.pyplot as plt and seaborn as sns.

Seaborn is a Python data visualization library based on matplotlib. Behind the scenes, seaborn uses matplotlib to draw its plots. When importing seaborn, it is recommended to call sns.set_theme() to apply the recommended seaborn visual style instead of the default matplotlib theme.

import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns

sns.set_theme()

Let's load this uniquely-formatted pokemon dataset.

pokemon = pd.read_csv("pokemon_viz.csv", index_col="Num")
pokemon

	Name	Type 1	Type 2	Total	HP	Attack	Defense	Sp. Atk	Sp. Def	Speed	Stage	Legendary
Num
1	Bulbasaur	Grass	Poison	318	45	49	49	65	65	45	1	False
2	Ivysaur	Grass	Poison	405	60	62	63	80	80	60	2	False
3	Venusaur	Grass	Poison	525	80	82	83	100	100	80	3	False
4	Charmander	Fire	NaN	309	39	52	43	60	50	65	1	False
5	Charmeleon	Fire	NaN	405	58	64	58	80	65	80	2	False
...	...	...	...	...	...	...	...	...	...	...	...	...
147	Dratini	Dragon	NaN	300	41	64	45	50	50	50	1	False
148	Dragonair	Dragon	NaN	420	61	84	65	70	70	70	2	False
149	Dragonite	Dragon	Flying	600	91	134	95	100	100	80	3	False
150	Mewtwo	Psychic	NaN	680	106	110	90	154	90	130	1	True
151	Mew	Psychic	NaN	600	100	100	100	100	100	100	1	False

151 rows × 12 columns

Figure-level versus axes-level functions

One way to draw a scatter plot comparing every pokemon's Attack and Defense stats is by calling sns.scatterplot. Because this plotting function has so many parameters, it's good practice to specify keyword arguments that tell Python which argument should go to which parameter.

sns.scatterplot(pokemon, x="Attack", y="Defense")

<Axes: xlabel='Attack', ylabel='Defense'>

The return type of sns.scatterplot is a matplotlib feature called axes that can be used to compose multiple plots into a single visualization. We can show two plots side-by-side by placing them on the same axes. For example, we could compare the attack and defense stats for two different groups of pokemon: not-Legendary and Legendary.

# Nested tuple unpacking!
fig, (ax1, ax2) = plt.subplots(nrows=1, ncols=2)
ax1.set_title("Not Legendary")
sns.scatterplot(pokemon[~pokemon["Legendary"]], x="Attack", y="Defense", ax=ax1)
ax2.set_title("Legendary")
sns.scatterplot(pokemon[pokemon["Legendary"]], x="Attack", y="Defense", ax=ax2)
fig.show()

Each problem in the plot above can be fixed manually by repeatedly editing and running the code until you get a satisfactory result, but it's a tedious process. Seaborn was invented to make our data visualization experience less tedious. Methods like sns.scatterplot are considered axes-level functions designed for interoperability with the rest of matplotlib, but they come at the cost of forcing you to deal with the tediousness of tweaking matplotlib.

Instead, the recommended way to create plots in seaborn is to use figure-level functions like sns.relplot as in relational plot. Figure-level functions return specialized seaborn objects (such as FacetGrid) that are intended to provide more usable results without tweaking.

sns.relplot(pokemon, x="Attack", y="Defense", col="Legendary")

<seaborn.axisgrid.FacetGrid at 0x7ca9dd87d450>

sns.displot(pokemon, x="Attack")

<seaborn.axisgrid.FacetGrid at 0x7ca9dd6b0c10>

By default, relational plots produce scatter plots but they can also produce line plots by specifying the keyword argument kind="line".

Alongside relplot, seaborn provides several other useful figure-level plotting functions:

• relplot for relational plots, such as scatter plots and line plots.
• displot for distribution plots, such as histograms and kernel density estimates.
• catplot for categorical plots, such as strip plots, box plots, violin plots, and bar plots.
• lmplot for relational plots with a regression fit, such as the scatter plot with regression fit below.

When reading documentation online, it is important to remember that we will only use figure-level plots in this course because they are the recommended approach. On the relative merits of figure-level functions in the seaborn documentation:

On balance, the figure-level functions add some additional complexity that can make things more confusing for beginners, but their distinct features give them additional power. The tutorial documentation mostly uses the figure-level functions, because they produce slightly cleaner plots, and we generally recommend their use for most applications. The one situation where they are not a good choice is when you need to make a complex, standalone figure that composes multiple different plot kinds. At this point, it’s recommended to set up the figure using matplotlib directly and to fill in the individual components using axes-level functions.

sns.lmplot(pokemon, x="Attack", y="Defense", col="Legendary")

<seaborn.axisgrid.FacetGrid at 0x7ca9dd41c460>

Customizing a FacetGrid plot

relplot, displot, catplot, and lmplot all return a FacetGrid, a specialized seaborn object that represents a data visualization canvas. As we've seen above, a FacetGrid can put two plots side-by-side and manage their axes by removing the y-axis labels on the right plot because they are the same as the plot on the left.

However, there are still many instances where we might want to customize a plot by changing labels or adding titles. We might want to create a bar plot to count the number of each type of pokemon.

sns.catplot(pokemon, x="Type 1", kind="count")

<seaborn.axisgrid.FacetGrid at 0x7ca9dd8293f0>

The pokemon types on the x-axis are hardly readable, the y-axis label "count" could use capitalization, and the plot could use a title. To modify the attributes of a plot, we can assign the returned FacetGrid to a variable like grid and then calling tick_params or set.

grid = sns.catplot(pokemon, x="Type 1", kind="count")
grid.tick_params(axis="x", rotation=60)
grid.set(title="Count of each primary pokemon type", xlabel="Primary Type", ylabel="Count")

<seaborn.axisgrid.FacetGrid at 0x7ca9dd41c460>

Practice: Life expectancy versus health expenditure

Seaborn includes a repository of example datasets that we can load into a DataFrame by calling sns.load_dataset. Let's examine the Life expectancy vs. health expenditure, 1970 to 2015 dataset that combines two data sources:

1. The Life expectancy at birth dataset from the UN World Population Prospects (2022): "For a given year, it represents the average lifespan for a hypothetical group of people, if they experienced the same age-specific death rates throughout their lives as the age-specific death rates seen in that particular year."
2. The Health expenditure (2010 int.-$) dataset from OECD.stat. "Per capita health expenditure and financing in OECD countries, measured in 2010 international dollars."

life_expectancy = sns.load_dataset("healthexp", index_col=["Year", "Country"])
life_expectancy

		Spending_USD	Life_Expectancy
Year	Country
1970	Germany	252.311	70.6
	France	192.143	72.2
	Great Britain	123.993	71.9
	Japan	150.437	72.0
	USA	326.961	70.9
...	...	...	...
2020	Germany	6938.983	81.1
	France	5468.418	82.3
	Great Britain	5018.700	80.4
	Japan	4665.641	84.7
	USA	11859.179	77.0

274 rows × 2 columns

Write a seaborn expression to create a line plot comparing the Year (x-axis) to the Life_Expectancy (y-axis) colored with hue="Country".

sns.lineplot(life_expectancy, x="Year", y="Life_Expectancy", hue="Country")

<Axes: xlabel='Year', ylabel='Life_Expectancy'>

sns.relplot(life_expectancy, x="Year", y="Life_Expectancy", hue="Country", kind="line")

<seaborn.axisgrid.FacetGrid at 0x7ca9dca3b640>

life_expectancy.loc[2019:2020, :]

		Spending_USD	Life_Expectancy
Year	Country
2019	Canada	5189.721	82.2
	Germany	6407.928	81.3
	France	5167.839	82.9
	Great Britain	4385.463	81.4
	Japan	4610.794	84.4
	USA	10855.517	78.8
2020	Canada	5828.324	81.7
	Germany	6938.983	81.1
	France	5468.418	82.3
	Great Britain	5018.700	80.4
	Japan	4665.641	84.7
	USA	11859.179	77.0

life_expectancy.loc[(2019:2020, "USA"), :]

  Cell In[14], line 1
    life_expectancy.loc[(2019:2020, "USA"), :]
                             ^
SyntaxError: invalid syntax

life_expectancy.loc[(slice(2019, 2020), "USA"), :]

		Spending_USD	Life_Expectancy
Year	Country
2019	USA	10855.517	78.8
2020	USA	11859.179	77.0

What makes bad figures bad?

In chapter 1 of Data Visualization, Kieran Hiely explains how data visualization is about communication and rhetoric.

While it is tempting to simply start laying down the law about what works and what doesn't, the process of making a really good or really useful graph cannot be boiled down to a list of simple rules to be followed without exception in all circumstances. The graphs you make are meant to be looked at by someone. The effectiveness of any particular graph is not just a matter of how it looks in the abstract, but also a question of who is looking at it, and why. An image intended for an audience of experts reading a professional journal may not be readily interpretable by the general public. A quick visualization of a dataset you are currently exploring might not be of much use to your peers or your students.

Bad taste

Kieran identifies three problems, the first of which is bad taste.

Kieran draws on Edward Tufte's principles (all quoted from Tufte 1983):

• have a properly chosen format and design
• use words, numbers, and drawing together
• display an accessible complexity of detail
• avoid content-free decoration, including chartjunk

In essence, these principles amount to "an encouragement to maximize the 'data-to-ink' ratio." In practice, our plotting libraries like seaborn do a fairly good job of providing defaults that generally follow these principles.

Bad data

The second problem is bad data, which can involve either cherry-picking data or presenting information in a misleading way.

In November of 2016, The New York Times reported on some research on people's confidence in the institutions of democracy. It had been published in an academic journal by the political scientist Yascha Mounk. The headline in the Times ran, "How Stable Are Democracies? ‘Warning Signs Are Flashing Red’” (Taub, 2016). The graph accompanying the article

This plot is one that is well-produced, and that we could reproduce by calling sns.relplot like we learned above. The x-axis shows the decade of birth for people all surveyed in the research study.

[But] scholars who knew the World Values Survey data underlying the graph noticed something else. The graph reads as though people were asked to say whether they thought it was essential to live in a democracy, and the results plotted show the percentage of respondents who said "Yes", presumably in contrast to those who said "No". But in fact the survey question asked respondents to rate the importance of living in a democracy on a ten point scale, with 1 being "Not at all Important" and 10 being "Absolutely Important". The graph showed the difference across ages of people who had given a score of "10" only, not changes in the average score on the question. As it turns out, while there is some variation by year of birth, most people in these countries tend to rate the importance of living in a democracy very highly, even if they do not all score it as "Absolutely Important". The political scientist Erik Voeten redrew the figure based using the average response.

Bad perception

The third problem is bad perception, which refers to how humans process the information contained in a visualization. Let's walk through section 1.3 on "Perception and data visualization".