Data visualization | MSc CSS: 07. Arranging plots

Iñaki Ucar

Source: _tutorials/07/07.Rmd

Compositions

Since the existence of the patchwork package, it is extremely simple to compose, align and annotate different ggplot graphs, and even merge them with other arbitrary elements. To demonstrate its capabilities, we will use the following plots:

p1 <- ggplot(mpg) +
  aes(displ, hwy) +
  geom_point() +
  labs(title="Plot 1")

p2 <- ggplot(mpg) +
  aes(as.character(year), fill=drv) +
  geom_bar(position="dodge") + 
  labs(title="Plot 2", x="year")

p3 <- ggplot(mpg) +
  aes(hwy, fill=drv) +
  geom_density(color=NA) + 
  facet_grid(drv~.) +
  labs(title="Plot 3")

p4 <- ggplot(mpg) +
  aes(drv, hwy) +
  stat_summary(aes(fill=drv), geom="col", fun.data=mean_se) +
  stat_summary(geom="errorbar", fun.data=mean_se, width=0.5) +
  labs(title="Plot 4")

Basic usage

Once loaded, this package extends the + operator to add plots together in a grid:

library(patchwork)

p1 + p2
p1 + p2 + p3
p1 + p2 + p3 + p4
p1 + plot_spacer() + p2 + plot_spacer() + p3 + plot_spacer()

For a finer control, it also provides the | (besides) and / (over) operators for plot stacking and packing:

p1 | p2
p1 / p2
(p1 / p2) | p3
p1 / (p2 | p3)

Note that patchwork automagically sets the proper margins for all the guides (axes, legends, titles…) so that all panels are perfectly aligned.

Controlling the layout

For even a richer control of the layout, we can just add the plots together and then add a plot_layout(), which has a bunch of options (see the help page) to control the grid:

p1 + p2 + p3 + p4 + 
  plot_layout(ncol=3)
p1 + p2 + p3 + p4 + 
  plot_layout(widths=c(2, 1))
p1 + p2 + p3 + p4 + 
  plot_layout(widths=c(2, 1), heights=unit(c(5, 1), c('cm', 'null')))

Moving beyond the grid, the design argument allows us to define textual representations of the layout:

layout <- "
#BB
ACD
#CD
"
p1 + p2 + p3 + p4 +
  plot_layout(design=layout)

layout <- "
AAB
C#B
CDD
"
p1 + p2 + p3 + p4 +
  plot_layout(design=layout)

Instead, a more programmatic approach uses the area() function for the same purpose:

layout <- c(
  area(1, 1, 1, 2),
  area(1, 3, 2, 3),
  area(2, 1, 3, 1),
  area(3, 2, 3, 3)
)
p1 + p2 + p3 + p4 +
  plot_layout(design=layout)

If the number of plots is not known beforehand, the wrap_plots() function makes it easy to take a list of plots and add them together into one composition, along with layout specifications.

plots <- list(p1, p2, p3, p4)
wrap_plots(plots, design=layout)

Checkpoint 1

Compose the four plots p1, p2, p3, p3 in a single column using only operators (+, /, |).
Make the same arrangement but using the plot_layout() function and without using the design argument.
Make the same arrangement but using the design argument.
Make the same arrangement but using the wrap_plots() function.

Fixed-aspect plots

Plots created with coord_fixed(), coord_polar() and coord_sf() cannot be resized to align with other plots. By default, patchwork aligns the other plots to them when possible.

p_fixed <- ggplot(mpg) +
  aes(hwy, cty) +
  geom_point() +
  coord_fixed() +
  labs(title="Fixed coords")

p_fixed + p1 + p2 + p3

But if we try to force the width, then the plot cannot be aligned.

p_fixed + p1 + p2 + p3 + plot_layout(widths=1)

Controlling the guides

When composing plots with multiple guides, it is often useful to collect all the guides and put them in the same place. Compare the following:

p1 + p2 + p3 + p4
p1 + p2 + p3 + p4 + plot_layout(guides="collect")

Note that guides were not only collected, but duplicates were removed. This option can be specified at several points for a finer control:

p1 + p2 + (p3 + plot_layout(guides="keep")) + p4 + plot_layout(guides="collect")

By default, guides are collected on the right of the composition, but a guide_area() can be specified as a place to put them:

layout <- "
AAB
CEB
CDD
"
p1 + p2 + p3 + p4 + guide_area() +
  plot_layout(design=layout, guides="collect")

Checkpoint 2

Take p2 and move the legend to the left in a standard way (modifying the theme).
Achieve the same result but using patchwork (use the widths argument to adjust the spacing).

Adding arbitrary elements

The wrap_elements() function can be used to wrap arbitrary grid grobs and add them to a composition.

text <- wrap_elements(grid::textGrob("Some really important text"))
table <- wrap_elements(gridExtra::tableGrob(mtcars[1:10, c("mpg", "disp")]))

p1 + table
text + p1

The ggplotify package provides functions for converting different graphics (base graphics, lattice graphics and others) to grobs, so that they can be composed too with ggplot objects.

Annotating the composition

The final composition can be annotated with titles, captions, tags, etc., via the plot_annotation() function:

p1 + p2 + p3 + p4 +
  plot_annotation(
    title = "Composition title",
    subtitle = "Composition subtitle",
    caption = "Composition caption",
    tag_levels = "a",
    tag_suffix = ")",
    theme = theme(plot.title=element_text(color="red"))
  )

Insets

Insets are essentially plot annotations, i.e., a plot on top of another plot. This can be achieved with multiple overlapping area() specifications, but the specific function inset_element() makes this task much easier. It allows us to freely position the inset relative either to the panel, plot of full area of the plot by specifying the left, bottom, right and top edges of the inset:

p1 + inset_element(p2, left=0.6, bottom=0.6, right=1, top=1, align_to="panel")
p1 + inset_element(p2, left=0.6, bottom=0.6, right=1, top=1, align_to="plot")
p1 + inset_element(p2, left=0.6, bottom=0.6, right=1, top=1, align_to="full")

And autotagging works as expected:

p1 + inset_element(p2, left=0.6, bottom=0.6, right=1, top=1, align_to="full") +
  plot_annotation(tag_levels="a")

One interesting use case of this technique is to zoom in a particular region:

xlim <- c(2.9, 3.2)
ylim <- c(21, 28)

p_zoom <- p1 +
  coord_cartesian(xlim=xlim, ylim=ylim, expand=FALSE) +
  labs(title=NULL) +
  theme_light() +
  theme(
    axis.title = element_blank(),
    plot.background = element_blank(),
    plot.margin = unit(c(0, 0, 0, 0), "mm")
  )

p1 +
  annotate("rect", xmin=xlim[1], xmax=xlim[2], ymin=ylim[1], ymax=ylim[2],
           fill=NA, color="black") +
  annotate("segment", x=xlim[2], xend=5.2, y=ylim[1], yend=29, linetype="dashed") +
  annotate("segment", x=xlim[1], xend=4.5, y=ylim[2], yend=45, linetype="dashed") +
  inset_element(p_zoom, left=0.5, bottom=0.45, right=0.65, top=0.95)

Checkpoint 3

Open the last plot in RStudio’s zoom panel. Play with the aspect ratio of the pop-up window. What is the issue with the segment specification above? How could this be improved? Investigate ggplot’s annotation_custom() for this task.

07. Arranging plots