This document contains all the code that is displayed during the workshop. The document is an RMarkdown document which means that it can be compiled, along with the code chunks thus executing and capturing the output of the code within the document. To read more about RMarkdown see the website for the package, as well as the Get Started guide.

Exercises

While it is encouraged to follow along in this document as the workshop progresses and execute the code to see the result, an important part is also to experiment and play, thus learning how the different settings affect the output.

The document will contain code chunks with the code examples discussed during the talk, but it will also contain chunks intended for completing small exercises. These will use the examples as a starting point and ask you to modify the code to achieve a given output. Completing these are optional, but highly recommended, either during or after the workshop.

Dependencies

This document comes with a list of required packages.

Datasets

We will use an assortment of datasets throughout the document. The purpose is mostly to showcase different plots, and less on getting some divine insight into the world. While not necessary we will call data(<dataset>) before using a new dataset to indicate the introduction of a new dataset.

Introduction

We will look at the basic ggplot2 use using the faithful dataset, giving information on the eruption pattern of the Old Faithful geyser in Yellowstone National Park.

data("faithful")
# Basic scatterplot
ggplot(data = faithful, 
       mapping = aes(x = eruptions, y = waiting)) + 
  geom_point()

# Data and mapping can be given both as global (in ggplot()) or per layer
ggplot() + 
  geom_point(mapping = aes(x = eruptions, y = waiting),
             data = faithful)

If an aesthetic is linked to data it is put into aes()

ggplot(faithful) + 
  geom_point(aes(x = eruptions, y = waiting, colour = eruptions < 3))

If you simple want to set it to a value, put it outside of aes()

ggplot(faithful) + 
  geom_point(aes(x = eruptions, y = waiting),
             colour = 'steelblue')

Some geoms only need a single mapping and will calculate the rest for you

ggplot(faithful) + 
  geom_histogram(aes(x = eruptions))
## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

geoms are drawn in the order they are added. The point layer is thus drawn on top of the density contours in the example below.

ggplot(faithful, aes(x = eruptions, y = waiting)) + 
  geom_density_2d() + 
  geom_point()

Exercise

Modify the code below to make the points larger squares and slightly transparent. See ?geom_point for more information on the point layer.

ggplot(faithful) + 
  geom_point(aes(x = eruptions, y = waiting))

Hint 1: transparency is controlled with alpha, and shape with shape Hint 2: rememberthe difference between mapping and setting aesthetics

Colour the two distributions in the histogram with different colours

ggplot(faithful) + 
  geom_histogram(aes(x = eruptions))
## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

Hint 1: For polygons you can map two different colour-like aesthetics: colour (the colour of the stroke) and fill (the fill colour)

Colour the distributions in the histogram by whether waiting is above or below 60. What happens?

ggplot(faithful) + 
  geom_histogram(aes(x = eruptions))
## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

Change the plot above by setting position = 'dodge' in geom_histogram() (while keeping the colouring by waiting). What do position control?

Add a line that separates the two point distributions. See ?geom_abline for how to draw straight lines from a slope and intercept.

ggplot(faithful) + 
  geom_point(aes(x = eruptions, y = waiting))

Stat

We will use the mpg dataset giving information about fuel economy on different car models.

Every geom has a stat. This is why new data (count) can appear when using geom_bar().

data("mpg")
ggplot(mpg) + 
  geom_bar(aes(x = class))

The stat can be overwritten. If we have precomputed count we don’t want any additional computations to perform and we use the identity stat to leave the data alone

library(dplyr)
## 
## Attaching package: 'dplyr'
## The following objects are masked from 'package:stats':
## 
##     filter, lag
## The following objects are masked from 'package:base':
## 
##     intersect, setdiff, setequal, union
mpg_counted <- mpg %>% 
  count(class, name = 'count')
ggplot(mpg_counted) + 
  geom_bar(aes(x = class, y = count), stat = 'identity')

Most obvious geom+stat combinations have a dedicated geom constructor. The one above is available directly as geom_col()

ggplot(mpg_counted) + 
  geom_col(aes(x = class, y = count))

Values calculated by the stat is available with the after_stat() function inside aes(). You can do all sorts of computations inside that.

ggplot(mpg) + 
  geom_bar(aes(x = class, y = after_stat(100 * count / sum(count))))

Many stats provide multiple variations of the same calculation, and provides a default (here, density)

ggplot(mpg) + 
  geom_density(aes(x = hwy))

While the others must be used with the after_stat() function

ggplot(mpg) + 
  geom_density(aes(x = hwy, y = after_stat(scaled)))

Exercises

While most people use geom_*() when adding layers, it is just as valid to add a stat_*() with an attached geom. Look at geom_bar() and figure out which stat it uses as default. Then modify the code to use the stat directly instead (i.e. adding stat_*() instead of geom_bar())

ggplot(mpg) + 
  geom_bar(aes(x = class))

Use stat_summary() to add a red dot at the mean hwy for each group

ggplot(mpg) + 
  geom_jitter(aes(x = class, y = hwy), width = 0.2)

Hint: You will need to change the default geom of stat_summary()

Scales

Scales define how the mapping you specify inside aes() should happen. All mappings have an associated scale even if not specified.

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, colour = class))

take control by adding one explicitly. All scales follow the same naming conventions.

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, colour = class)) + 
  scale_colour_brewer(type = 'qual')

Positional mappings (x and y) also have associated scales.

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy)) + 
  scale_x_continuous(breaks = c(3, 5, 6)) + 
  scale_y_continuous(trans = 'log10')

Exercises

Use RColorBrewer::display.brewer.all() to see all the different palettes from Color Brewer and pick your favourite. Modify the code below to use it

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, colour = class)) + 
  scale_colour_brewer(type = 'qual')

Modify the code below to create a bubble chart (scatterplot with size mapped to a continuous variable) showing cyl with size. Make sure that only the present amount of cylinders (4, 5, 6, and 8) are present in the legend.

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, colour = class)) + 
  scale_colour_brewer(type = 'qual')

Hint: The breaks argument in the scale is used to control which values are present in the legend.

Explore the different types of size scales available in ggplot2. Is the default the most appropriate here?

Modify the code below so that colour is no longer mapped to the discrete class variable, but to the continuous cty variable. What happens to the guide?

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, colour = class, size = cty))

The type of guide can be controlled with the guide argument in the scale, or with the guides() function. Continuous colours have a gradient colour bar by default, but setting it to legend will turn it back to the standard look. What happens when multiple aesthetics are mapped to the same variable and uses the guide type?

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, colour = cty, size = cty))

Facets

The facet defines how data is split among panels. The default facet (facet_null()) puts all the data in a single panel, while facet_wrap() and facet_grid() allows you to specify different types of small multiples

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy)) + 
  facet_wrap(~ class)

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy)) + 
  facet_grid(year ~ drv)

Exercises

One of the great things about facets is that they share the axes between the different panels. Sometimes this is undiserable though, and the behaviour can be changed with the scales argument. Experiment with the different possible settings in the plot below:

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy)) + 
  facet_wrap(~ drv)

Usually the space occupied by each panel is equal. This can create problems when different scales are used. Modify the code below so that the y scale differs between the panels in the plot. What happens?

ggplot(mpg) + 
  geom_bar(aes(y = manufacturer)) + 
  facet_grid(class ~ .)

Use the space argument in facet_grid() to change the plot above so each bar has the same width again.

Facets can be based on multiple variables by adding them together. Try to recreate the same panels present in the plot below by using facet_wrap()

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy)) + 
  facet_grid(year ~ drv)

Coordinates

The coordinate system is the fabric you draw your layers on in the end. The default `coord_cartesion provides the standard rectangular x-y coordinate system. Changing the coordinate system can have dramatic effects

ggplot(mpg) + 
  geom_bar(aes(x = class)) + 
  coord_polar()

ggplot(mpg) + 
  geom_bar(aes(x = class)) + 
  coord_polar(theta = 'y') + 
  expand_limits(y = 70)

You can zoom both on the scale…

ggplot(mpg) + 
  geom_bar(aes(x = class)) + 
  scale_y_continuous(limits = c(0, 40))
## Warning: Removed 3 rows containing missing values (geom_bar).

and in the coord. You usually want the latter as it avoids changing the plottet data

ggplot(mpg) + 
  geom_bar(aes(x = class)) + 
  coord_cartesian(ylim = c(0, 40))

Exercises

In the same way as limits can be set in both the positional scale and the coord, so can transformations, using coord_trans(). Modify the code below to apply a log transformation to the y axis; first using scale_y_continuous(), and then using coord_trans(). Compare the results — how do they differ?

ggplot(mpg) + 
  geom_point(aes(x = hwy, y = displ))

Coordinate systems are particularly important in cartography. While we will not spend a lot of time with it in this workshop, spatial plotting is well supported in ggplot2 with geom_sf() and coord_sf() (which interfaces with the sf package). The code below produces a world map. Try changing the crs argument in coord_sf() to be '+proj=robin' (This means using the Robinson projection).

# Get the borders of all countries
world <- sf::st_as_sf(maps::map('world', plot = FALSE, fill = TRUE))
world <- sf::st_wrap_dateline(world, 
                              options = c("WRAPDATELINE=YES", "DATELINEOFFSET=180"),
                              quiet = TRUE)
## Warning in CPL_wrap_dateline(st_geometry(x), options, quiet): GDAL Error 6: GEOS
## support not enabled.

## Warning in CPL_wrap_dateline(st_geometry(x), options, quiet): GDAL Error 6: GEOS
## support not enabled.

## Warning in CPL_wrap_dateline(st_geometry(x), options, quiet): GDAL Error 6: GEOS
## support not enabled.

## Warning in CPL_wrap_dateline(st_geometry(x), options, quiet): GDAL Error 6: GEOS
## support not enabled.

## Warning in CPL_wrap_dateline(st_geometry(x), options, quiet): GDAL Error 6: GEOS
## support not enabled.

## Warning in CPL_wrap_dateline(st_geometry(x), options, quiet): GDAL Error 6: GEOS
## support not enabled.
# Plot code
ggplot(world) + 
  geom_sf() + 
  coord_sf(crs = "+proj=moll")

Maps are a huge area in data visualisation and simply too big to cover in this workshop. If you want to explore further I advice you to explore the r-spatial wbsite as well as the website for the sf package

Theme

Theming defines the feel and look of your final visualisation and is something you will normally defer to the final polishing of the plot. It is very easy to change looks with a prebuild theme

ggplot(mpg) + 
  geom_bar(aes(y = class)) + 
  facet_wrap(~year) + 
  theme_minimal()

Further adjustments can be done in the end to get exactly the look you want

ggplot(mpg) + 
  geom_bar(aes(y = class)) + 
  facet_wrap(~year) + 
  labs(title = "Number of car models per class",
       caption = "source: http://fueleconomy.gov",
       x = NULL,
       y = NULL) +
  scale_x_continuous(expand = c(0, NA)) + 
  theme_minimal() + 
  theme(
    text = element_text('Avenir Next Condensed'),
    strip.text = element_text(face = 'bold', hjust = 0),
    plot.caption = element_text(face = 'italic'),
    panel.grid.major = element_line('white', size = 0.5),
    panel.grid.minor = element_blank(),
    panel.grid.major.y = element_blank(),
    panel.ontop = TRUE
  )

Exercises

Themes can be overwhelming, especially as you often try to optimise for beauty while you learn. To remove the last part of the equation, the exercise is to take the plot given below and make it as hideous as possible using the theme function. Go absolutely crazy, but take note of the effect as you change different settings.

ggplot(mpg) + 
  geom_bar(aes(y = class, fill = drv)) + 
  facet_wrap(~year) + 
  labs(title = "Number of car models per class",
       caption = "source: http://fueleconomy.gov",
       x = 'Number of cars',
       y = NULL)

Extensions

While ggplot2 comes with a lot of batteries included, the extension ecosystem provides priceless additinal features

Plot composition

We start by creating 3 separate plots

p1 <- ggplot(msleep) + 
  geom_boxplot(aes(x = sleep_total, y = vore, fill = vore))
p1

p2 <- ggplot(msleep) + 
  geom_bar(aes(y = vore, fill = vore))
p2

p3 <- ggplot(msleep) + 
  geom_point(aes(x = bodywt, y = sleep_total, colour = vore)) + 
  scale_x_log10()
p3

Combining them with patchwork is a breeze using the different operators

library(patchwork)
p1 + p2 + p3

(p1 | p2) / 
   p3

p_all <- (p1 | p2) / 
            p3
p_all + plot_layout(guides = 'collect')

p_all & theme(legend.position = 'none')

p_all <- p_all & theme(legend.position = 'none')
p_all + plot_annotation(
  title = 'Mammalian sleep patterns',
  tag_levels = 'A'
)

Excercises

Patchwork will assign the same amount of space to each plot by default, but this can be controlled with the widths and heights argument in plot_layout(). This can take a numeric vector giving their relative sizes (e.g. c(2, 1) will make the first plot twice as big as the second). Modify the code below so that the middle plot takes up half of the total space:

p <- ggplot(mtcars) + 
  geom_point(aes(x = disp, y = mpg))
p + p + p

The & operator can be used with any type of ggplot2 object, not just themes. Modify the code below so the two plots share the same y-axis (same limits)

p1 <- ggplot(mtcars[mtcars$gear == 3,]) + 
  geom_point(aes(x = disp, y = mpg))
p2 <- ggplot(mtcars[mtcars$gear == 4,]) + 
  geom_point(aes(x = disp, y = mpg))
p1 + p2

Patchwork contains many features for fine tuning the layout and annotation. Very complex layouts can be obtained by providing a design specification to the design argument in plot_layout(). The design can be defined as a textual representation of the cells. Use the layout given below. How should the textual representation be undertood.

p1 <- ggplot(mtcars) + 
  geom_point(aes(x = disp, y = mpg))
p2 <- ggplot(mtcars) + 
  geom_bar(aes(x = factor(gear)))
p3 <- ggplot(mtcars) + 
  geom_boxplot(aes(x = factor(gear), y = mpg))
layout <- '
AA#
#BB
C##
'
p1 + p2 + p3 + plot_layout(design = layout)

Animation

ggplot2 is usually focused on static plots, but gganimate extends the API and grammar to describe animations. As such it feels like a very natural extension of using ggplot2

ggplot(economics) + 
  geom_line(aes(x = date, y = unemploy))

library(gganimate)
## No renderer backend detected. gganimate will default to writing frames to separate files
## Consider installing:
## - the `gifski` package for gif output
## - the `av` package for video output
## and restarting the R session
ggplot(economics) + 
  geom_line(aes(x = date, y = unemploy)) + 
  transition_reveal(along = date)
## Warning: No renderer available. Please install the gifski, av, or magick package
## to create animated output
## NULL

There are many different transitions that control how data is interpreted for animation, as well as a range of other animation specific features

ggplot(mpg) + 
  geom_bar(aes(x = factor(cyl)))

ggplot(mpg) + 
  geom_bar(aes(x = factor(cyl))) + 
  labs(title = 'Number of cars in {closest_state} by number of cylinders') + 
  transition_states(states = year) + 
  enter_grow() + 
  exit_fade()
## Warning: No renderer available. Please install the gifski, av, or magick package
## to create animated output
## NULL

Exercises

The animation below will animate between points showing cars with different cylinders.

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy)) + 
  ggtitle("Cars with {closest_state} cylinders") + 
  transition_states(factor(cyl))
## Warning: No renderer available. Please install the gifski, av, or magick package
## to create animated output
## NULL

gganimate uses the group aesthetic to match observations between states. By default the group aesthetic is set to the same value, so observations are matched by their position (first row of 4 cyl is matched to first row of 5 cyl etc.). This is clearly wrong here (why?). Add a mapping to the group aesthetic to ensure that points do not move between the different states.

In the presence of discrete aesthetic mappings (colour below), the group is deduced if not given. The default behaviour of objects that appear and disappear during the animation is to simply pop in and out of existance. enter_*() and exit_*() functions can be used to control this behaviour. Experiment with the different enter and exit functions provided by gganimate below. What happens if you add multiple enter or exit functions to the same animation?

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, colour = factor(cyl))) + 
  ggtitle("Cars with {closest_state} cylinders") + 
  transition_states(factor(cyl))
## Warning: No renderer available. Please install the gifski, av, or magick package
## to create animated output
## NULL

In the animation below (as in all the other animations) the changes happens at constant speed. How values change during an animation is called easing and can be controlled using the ease_aes() function. Read the documentation for ease_aes() and experiment with different easings in the animation.

mpg2 <- tidyr::pivot_longer(mpg, c(cty,hwy))
ggplot(mpg2) + 
  geom_point(aes(x = displ, y = value)) + 
  ggtitle("{if (closest_state == 'cty') 'Efficiency in city' else 'Efficiency on highway'}") + 
  transition_states(name)
## Warning: No renderer available. Please install the gifski, av, or magick package
## to create animated output
## NULL

Annotation

Text is a huge part of storytelling with your visualisation. Historically, textual annotations has not been the best part of ggplot2 but new extensions make up for that.

Standard geom_text will often result in overlaping labels

ggplot(mtcars, aes(x = disp, y = mpg)) + 
  geom_point() + 
  geom_text(aes(label = row.names(mtcars)))

ggrepel takes care of that

library(ggrepel)
ggplot(mtcars, aes(x = disp, y = mpg)) + 
  geom_point() + 
  geom_text_repel(aes(label = row.names(mtcars)))
## Warning: ggrepel: 1 unlabeled data points (too many overlaps). Consider
## increasing max.overlaps

If you want to highlight certain parts of your data and describe it, the geom_mark_*() family of geoms have your back

library(ggforce)
ggplot(mtcars, aes(x = disp, y = mpg)) +
  geom_point() + 
  geom_mark_ellipse(aes(filter = gear == 4,
                        label = '4 gear cars',
                        description = 'Cars with fewer gears tend to both have higher yield and lower displacement'))

Exercises

ggrepel has a tonne of settings for controlling how text labels move. Often, though, the most effective is simply to not label everything. There are two strategies for that: Either only use a subset of the data for the repel layer, or setting the label to "" for those you don’t want to plot. Try both in the plot below where you only label 10 random points.

mtcars2 <- mtcars
mtcars2$label <- rownames(mtcars2)
points_to_label <- sample(nrow(mtcars), 10)
ggplot(mtcars2, aes(x = disp, y = mpg)) + 
  geom_point() + 
  geom_text_repel(aes(label = label))
## Warning: ggrepel: 1 unlabeled data points (too many overlaps). Consider
## increasing max.overlaps

Explore the documentation for geom_text_repel. Find a way to ensure that the labels in the plot below only repels in the vertical direction

mtcars2$label <- ""
mtcars2$label[1:10] <- rownames(mtcars2)[1:10]
ggplot(mtcars2, aes(x = disp, y = mpg)) + 
  geom_point() + 
  geom_text_repel(aes(label = label))

ggforce comes with 4 different types of mark geoms. Try them all out in the code below:

ggplot(mtcars, aes(x = disp, y = mpg)) +
  geom_point() + 
  geom_mark_ellipse(aes(filter = gear == 4,
                        label = '4 gear cars'))

Networks

ggplot2 has been focused on tabular data. Network data in any shape and form is handled by ggraph

library(ggraph)
library(tidygraph)
## 
## Attaching package: 'tidygraph'
## The following object is masked from 'package:stats':
## 
##     filter
graph <- create_notable('zachary') %>% 
  mutate(clique = as.factor(group_infomap()))
ggraph(graph) + 
  geom_mark_hull(aes(x, y, fill = clique)) + 
  geom_edge_link() + 
  geom_node_point(size = 2)
## Using `stress` as default layout
## Warning: The concaveman package is required for geom_mark_hull

dendrograms are just a specific type of network

iris_clust <- hclust(dist(iris[, 1:4]))
ggraph(iris_clust) + 
  geom_edge_bend() + 
  geom_node_point(aes(filter = leaf))
## Using `dendrogram` as default layout

Exercies

Most network plots are defined by a layout algorithm, which takes the network structure and calculate a position for each node. The layout algorithm is global and set in the ggraph(). The default auto layout will inspect the network object and try to choose a sensible layout for it (e.g. dendrogram for a hierarchical clustering as above). There is, however no optimal layout and it is often a good idea to try out different layouts. Try out different layouts in the graph below. See the the website for an overview of the different layouts.

ggraph(graph) + 
  geom_edge_link() + 
  geom_node_point(aes(colour = clique), size = 3)
## Using `stress` as default layout

There are many different ways to draw edges. Try to use geom_edge_parallel() in the graph below to show the presence of multiple edges

highschool_gr <- as_tbl_graph(highschool)
ggraph(highschool_gr) + 
  geom_edge_link() + 
  geom_node_point()
## Using `stress` as default layout

Faceting works in ggraph as it does in ggplot2, but you must choose to facet by either nodes or edges. Modify the graph below to facet the edges by the year variable (using facet_edges())

ggraph(highschool_gr) + 
  geom_edge_fan() + 
  geom_node_point()
## Using `stress` as default layout

Looks

Many people have already desgned beautiful (and horrible) themes for you. Use them as a base

p <- ggplot(mtcars, aes(mpg, wt)) +
  geom_point(aes(color = factor(carb))) +
  labs(
    x = 'Fuel efficiency (mpg)', 
    y = 'Weight (tons)',
    title = 'Seminal ggplot2 example',
    subtitle = 'A plot to show off different themes',
    caption = 'Source: It’s mtcars — everyone uses it'
  )
library(hrbrthemes)
## NOTE: Either Arial Narrow or Roboto Condensed fonts are required to use these themes.
##       Please use hrbrthemes::import_roboto_condensed() to install Roboto Condensed and
##       if Arial Narrow is not on your system, please see https://bit.ly/arialnarrow
p + 
  scale_colour_ipsum() + 
  theme_ipsum()

library(ggthemes)
p + 
  scale_colour_excel() + 
  theme_excel()

Drawing anything

states <- c(
  'eaten', "eaten but said you didn\'t", 'cat took it', 'for tonight',
  'will decompose slowly'
)
pie <- data.frame(
  state = factor(states, levels = states),
  amount = c(4, 3, 1, 1.5, 6),
  stringsAsFactors = FALSE
)
ggplot(pie) + 
  geom_col(aes(x = 0, y = amount, fill = state))

ggplot(pie) + 
  geom_col(aes(x = 0, y = amount, fill = state)) + 
  coord_polar(theta = 'y')

ggplot(pie) + 
  geom_col(aes(x = 0, y = amount, fill = state)) + 
  coord_polar(theta = 'y') + 
  scale_fill_tableau(name = NULL,
                     guide = guide_legend(ncol = 2)) + 
  theme_void() + 
  theme(legend.position = 'top', 
        legend.justification = 'left')

ggplot(pie) + 
  geom_arc_bar(aes(x0 = 0, y0 = 0, r0 = 0, r = 1, amount = amount, fill = state), stat = 'pie') + 
  coord_fixed()

ggplot(pie) + 
  geom_arc_bar(aes(x0 = 0, y0 = 0, r0 = 0, r = 1, amount = amount, fill = state), stat = 'pie') + 
  coord_fixed() + 
  scale_fill_tableau(name = NULL,
                     guide = guide_legend(ncol = 2)) + 
  theme_void() + 
  theme(legend.position = 'top', 
        legend.justification = 'left')

ggplot(mpg) + 
  # geom_bar(aes(x = hwy), stat = 'bin')
  geom_histogram(aes(x = hwy))
## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

ggplot(mpg) + 
  geom_bar(aes(x = hwy)) + 
  scale_x_binned(n.breaks = 30, guide = guide_axis(n.dodge = 2))