Grammar of Graphics in R

Let’s get to coding!

We will start with an example where we take mpg data and make a scatterplot.

How do we get from this:

The data

head(mpg)

to this:

The plot

Let’s plot!

Load the ggplot2 library:

library(ggplot2)

ggplot is the base layer.

ggplot()

This is a good place to for us to define our plot’s default data.

ggplot(data = mpg)

There isn’t much being plotted yet… Let’s tell our plot about our mapping, where our x will be our cty and y will be our hwy.

ggplot(
    data = mpg,
    mapping = aes(x = cty, y = hwy)
  )

Our plot now has enough information to set some automatic axis limits based on the data.

ggplot(
    data = mpg,
    mapping = aes(x = cty, y = hwy)
  ) +
  geom_point()

We can map the number of cylinders as a set of discrete values to color.

ggplot(
    data = mpg,
    mapping = aes(x = cty, y = hwy)
  ) +
  geom_point(
    mapping = aes(color = factor(cyl))
  )

At this point, we have the base of the plot. We can save this to a variable, mpg_base_plot, and make additional adjustments.

mpg_base_plot <- ggplot(
    data = mpg,
    mapping = aes(x = cty, y = hwy)
  ) +
  geom_point(
    mapping = aes(color = factor(cyl))
  )
mpg_base_plot

To add the reference line, we add a layer like so:

mpg_base_plot +
  geom_abline(
    slope = 1,
    intercept = 0,
    color = 'black',
    linetype = 'dotted'
  )

Note that this geom layer can be added on top of a blank plot. All geom_ prefixed functions return an environment object that holds onto the information we call it with.

ggplot() +
  geom_abline(
    slope = 1,
    intercept = 0,
    color = 'black',
    linetype = 'dotted'
  )

Let’s continue. Our plot looks a bit funny – the grid for the x and y are not 1 to 1. Let’s use a coord_ function to adjust this:

mpg_base_plot +
  geom_abline(
    slope = 1,
    intercept = 0,
    color = 'black',
    linetype = 'dotted'
  ) +
  coord_fixed(ratio = 1)

We can clean up the plot like so:

mpg_plot <- mpg_base_plot +
  geom_abline(
    slope = 1,
    intercept = 0,
    color = 'black',
    linetype = 'dotted'
  ) +
  coord_fixed(ratio = 1) +
  scale_color_manual(
    name = '# of Cyl',
    values = c('skyblue','royalblue', 'blue', 'navy')
  ) +
  ggtitle('Miles per gallon') +
  theme_light()
mpg_plot

We can further manipulate the plot using things like facet_.

mpg_plot +
  facet_grid(.~class)

Let’s experiment

We can get the base plots from the ggplot cheatsheets on our computer by sourcing this R script.

source('https://goo.gl/3EQSXt')

The magic behind the +

We can discover how the + is helping us construct these plots by running:

methods('+')

[1] +.Date   +.gg*    +.POSIXt
see '?methods' for accessing help and source code

From the result, we can see that the + function is overloaded for objects that match the gg namespace. If we look at the code for this function, we learn more about how it works.

Object-oriented

We can explore more of how ggplot works underneath by looking at:

base_plot <- ggplot()
names(base_plot)

[1] "data"        "layers"      "scales"      "mapping"     "theme"       "coordinates" "facet"       "plot_env"   
[9] "labels"     

It’s an object that can be mutated. Other functions output objects as well.

mpg_mapping <- aes(x = cty, y = hwy)
mpg_mapping

* x -> cty
* y -> hwy

Let’s mutate the base_plot:

base_plot$data <- mpg
base_plot$mapping <- mpg_mapping
base_plot$layers <- c(base_plot$layers, geom_point())
base_plot

Extensibility

You can also create custom stat_s, geom_s, and theme_s as needed. This vignette explains the object-oriented patterns for ggplot2, and how to extend the library for your needs.

There’s actually a lot we can learn from the source code, such as all possible aes parameters.

The makings of a pie

Grammar-based graphics lends flexibility

Back to Leland Wilkinson’s point about making a pie – that a pie chart is really a stacked bar chart that has been transformed with polar coordinates where the y is mapped to the angle, a.k.a. theta.

We start with a bar chart:

diamonds_by_cut_base <- ggplot(data = diamonds, mapping = aes(x = cut, fill = cut))
diamonds_by_cut_base +
  geom_bar()

Here, we have a stacked bar chart of diamonds grouped by cut.

diamonds_by_cut_stacked <- diamonds_by_cut_base +
  geom_bar(width = 1, aes(x=factor("")))
diamonds_by_cut_stacked

Taking the bar chart, we simply “add” a coord_polar to it.

diamonds_by_cut_stacked +
  coord_polar()  +
  scale_x_discrete("")

By default, coord_polar maps the theta from x and the radius from y.

bullseye_coords <- coord_polar()
bullseye_coords$theta

[1] "x"

bullseye_coords$r

[1] "y"

We need to a coord_polar where the theta maps from y:

pie_coords <- coord_polar(theta = "y")
pie_coords$theta

[1] "y"

Now, we can transform the stacked bar chart to a pie chart:

diamonds_by_cut_stacked +
  pie_coords

?coord_polar

A layered grammar is powerful

The full grammar consists of:

• The base plot – ggplot – with default data and mappings
• Any number of layers – geom_ or stat_ – each with
  • data
    • plot’s data by default
  • mapping
    • plot’s mappings by default
    • defined by aesthetics
  • geom_
    • Sometimes defined by default by stat_ if defined
  • stat_
    • Identity by default
    • Sometimes defined by default by geom_ if defined
  • position
    • Optional, any adjustments necessary to geom positioning

Additionally, the grammar includes control over the whole plot’s

• scales
  • to define how data values map aesthetic values
• coordinates
• facets
  • subplots

Reading and Sources

A Layered Grammar of Graphics

Mastering the Grammar

R for Data Science: Visualizing Data

Extending ggplot2

The R Graph Gallery

Data Camp: Data Visualization with ggplot2

Swirl: Exploratory Data Analysis

ggplot2 Cheatsheet

Telling stories with data using the grammar of graphics

Introduction to R Graphics with ggplot2

ggplot2 source code

ggplot2: Elegant Graphics for Data Analysis

The Grammar of Graphics

Semiology of Graphics