Geometries
Three of the geometries described below allow the addition of plot layers containing insets. Insets can be plots, tables, bitmaps, or grid objects. Insets can be also added as annotations. Using for data a tibble with a list column containing data frames or tibbles allows like any other geom, the use of grouping, multiple insets per panel, faceting with different tables per panel, and different number of insets in each panel, i.e., individual tables added to a plot with geom_table behave similarly to individual character values added with geom_text.
Other geometries also described in this section support the use of native plot coordinates for positioning elements in the plotting area. Obviously these geometries are not meant to be used to display data, but instead they make it possible to add annotations to plots consistently across data sets and varying scales used.
geom_table() and stat_fmt_table()
The geometry geom_table() plots a data frame or tibble, nested in a tibble passed as data argument, using aesthetics x and y for positioning, and label for the list of data frames containing the data for the tables. The tables are created as a ‘grid’ grobs and added as usual to the ggplot object. In contrast to “standard” geoms, this geom by default does not inherit the globally mapped aesthetics.
tb <- mpg %>%
group_by(cyl) %>%
summarise(hwy = median(hwy), cty = median(cty))
data.tb <- tibble(x = 7, y = 44, tb = list(tb))
ggplot(mpg, aes(displ, hwy, colour = factor(cyl))) +
geom_table(data = data.tb, aes(x, y, label = tb)) +
geom_point()
In plots with a single panel it is easier to use annotate() to add inset tables, giving the same plot as above. In this case single data frames, ggplots or grobs do not need to be wrapped in a list, although lists are also supported.
tb <- mpg %>%
group_by(cyl) %>%
summarise(hwy = median(hwy), cty = median(cty))
ggplot(mpg, aes(displ, hwy, colour = factor(cyl))) +
annotate("table", x = 7, y = 44, label = tb) +
geom_point()
Table themes are supported through parameter table.theme and if variables or constants are mapped to the colour, fill, size, family, aesthetics they override the corresponding theme settings. The display of rownames and colnames can be enabled or disable through parameter table.rownames and table.colnames and the horizontal justification of text in the core of the table through parameter table.hjust.
Parameter table.theme accepts as arguments NULL for use of the current default, a ttheme constructor function such as those defined in package ‘gridExtra’, or the variations on them defined in this package. The current default can be set with function ttheme_set().
ggplot(mpg, aes(displ, hwy, colour = factor(cyl))) +
geom_table(data = data.tb, aes(x, y, label = tb),
table.theme = ttheme_gtsimple,
table.hjust = 0, colour = "darkred", fill = "#FFFFBB") +
geom_point()
Using stat_fmt_tb() we can rename columns and rows of the tibble, reorder them and/or select a subset of columns or rows as shown below. To provide a complete example we also replace the names of the scales for x, y and color aesthetics. Here we pass a character vector with the original names of the columns in full, but partial matching is tried when needed. It is also possible to use a numeric vector of positional indexes.
ggplot(mpg, aes(displ, hwy, colour = factor(cyl))) +
geom_table(data = data.tb, aes(x, y, label = tb),
table.theme = ttheme_gtlight,
size = 3, colour = "darkblue",
stat = "fmt_tb",
tb.vars = c(Cylinders = "cyl", MPG = "hwy"), # rename
tb.rows = 4:1) + # change order
labs(x = "Engine displacement (l)", y = "Fuel use efficiency (MPG)",
colour = "Engine cylinders\n(number)") +
geom_point() +
theme_bw()
Parsed text, using plot math syntax is supported in the table, with fall-back to plain text in case of parsing errors, on a cell by cell basis. Here we plot the MPG for city traffic and we can see that the plotting area expands to include the coordinates at which the table is anchored. Justification is by default set to "inward" which ensures that the table is fully within the plotting region.
tb.pm <- tibble(Parameter = c("frac(beta[1], a^2)", "frac(beta[2], a^3)"),
Value = c("10^2.4", "10^3.532"))
data.tb <- tibble(x = 7, y = 44, tb = list(tb.pm))
ggplot(mpg, aes(displ, cty)) +
geom_point() +
geom_table(data = data.tb, aes(x, y, label = tb), parse = TRUE) +
theme_bw()
As implemented, there is no limitation to the number of insets, and faceting is respected. If the base plot shows a map, multiple small tables could be superimposed on different countries or regions. The size of the insets is set relative to the main plot, so the combined plot can be scaled.
Please see section Normalised Parent Coordinates below for a description of geom_table_npc().
geom_plot()
The geom_plot() geometry plots ggplot objects, nested in a tibble passed as data argument, using aesthetics x and y for positioning, and label for the ggplot object containing the definition of the plot to be nested. With this approach in plots with facets the insets can be different in each panel. It is also possible to inset more than one plot in a single call simply by creating a tibble with multiple rows.
Behind the scenes, one Grob is created for each plot to be inset. The conversion is done with ggplotGrob() and the Grobs added the main ggplot object.
As an example we produce a plot where the inset plot is a zoomed-in detail from the main plot. In this case the main and inset plots start as the same plot. In most cases the size of text and other elements in the inset should be smaller than in the main plot. Here we override the default theme setting the base_size from its default of 11 pt to 8 pt.
p <- ggplot(mpg, aes(displ, hwy, colour = factor(cyl))) +
geom_point()
data.tb <-
tibble(x = 7, y = 44,
plot = list(p +
coord_cartesian(xlim = c(4.9, 6.2),
ylim = c(13, 21)) +
labs(x = NULL, y = NULL) +
theme_bw(8) +
scale_colour_discrete(guide = FALSE)))
ggplot(mpg, aes(displ, hwy, colour = factor(cyl))) +
geom_plot(data = data.tb, aes(x, y, label = plot)) +
annotate(geom = "rect",
xmin = 4.9, xmax = 6.2, ymin = 13, ymax = 21,
linetype = "dotted", fill = NA, colour = "black") +
geom_point()
In general, the inset plot can be any ggplot object, allowing the creation of very different combinations of main plot and inset plots. Here we use the inset to show summaries as in the previous example of an inset table.
p <- ggplot(mpg, aes(factor(cyl), hwy, fill = factor(cyl))) +
stat_summary(geom = "col", fun = mean, width = 2/3) +
labs(x = "Number of cylinders", y = NULL, title = "Means") +
scale_fill_discrete(guide = FALSE)
data.tb <- tibble(x = 7, y = 44,
plot = list(p +
theme_bw(8)))
ggplot(mpg, aes(displ, hwy, colour = factor(cyl))) +
geom_plot(data = data.tb, aes(x, y, label = plot)) +
geom_point() +
labs(x = "Engine displacement (l)", y = "Fuel use efficiency (MPG)",
colour = "Engine cylinders\n(number)") +
theme_bw()
The same plot as above can be created using annotate() but be aware that when using facets in ‘ggplot2’ annotations are identical in all panels.
p <- ggplot(mpg, aes(factor(cyl), hwy, fill = factor(cyl))) +
stat_summary(geom = "col", fun = mean, width = 2/3) +
labs(x = "Number of cylinders", y = NULL, title = "Means") +
scale_fill_discrete(guide = FALSE)
ggplot(mpg, aes(displ, hwy, colour = factor(cyl))) +
annotate("plot", x = 7, y = 44, label = p + theme_bw(8)) +
geom_point() +
labs(x = "Engine displacement (l)", y = "Fuel use efficiency (MPG)",
colour = "Engine cylinders\n(number)") +
theme_bw()
As implemented, there is no limitation to the number of insets, and faceting is respected. If the base plot shows a map or a bitmap, multiple small plots could be superimposed on different countries or regions. The size of the insets is set relative to the main plot, so the combined plot can be scaled. A possible unintuitive but useful feature, is that the theme is linked to each plot.
Please see section Normalised Parent Coordinates below for a description of geom_plot_npc().
geom_grob()
The geom_grob() geometry plots grobs (graphical objects as created with ‘grid’), nested in a tibble passed as data argument, using aesthetics x and y for positioning, and label for the Grob object. While geom_table() and geom_plot() take as values to plot tibbles or data frames, and ggplots, respectively, and convert them into Grobs before adding them to the plot, geom_grob() expects Grobs ready to be rendered.
file.name <-
system.file("extdata", "Isoquercitin.png",
package = "ggpmisc", mustWork = TRUE)
Isoquercitin <- magick::image_read(file.name)
grobs.tb <- tibble(x = c(0, 10, 20, 40), y = c(4, 5, 6, 9),
width = c(0.05, 0.05, 0.01, 1),
height = c(0.05, 0.05, 0.01, 0.3),
grob = list(grid::circleGrob(),
grid::rectGrob(),
grid::textGrob("I am a Grob"),
grid::rasterGrob(image = Isoquercitin)))
ggplot() +
geom_grob(data = grobs.tb,
aes(x, y, label = grob, vp.width = width, vp.height = height),
hjust = 0.7, vjust = 0.55) +
scale_y_continuous(expand = expansion(mult = 0.3, add = 0)) +
scale_x_continuous(expand = expansion(mult = 0.2, add = 0)) +
theme_bw(12)
As shown above for inset tables and inset plots, it is also possible to use annotate() with Grobs. The next example insets a single Grob. Here we reuse the bitmap Isoquercitin read in the previous example. The Grob is contained in a viewport. Here setting width = 1 (“npc” units) when creating the Grob from the bitmap ensures that the bitmap fills the width of the viewport (to ensure that the inset is not distorted, set only one of width or height). The argument to vp.width or vp.height, also in “npc” units, determines the size of the Grob relative to the size of the plotting area.
ggplot() +
annotate("grob", x = 1, y = 3, vp.width = 0.5,
label = grid::rasterGrob(image = Isoquercitin, width = 1)) +
theme_bw(12)
geom_grob() is designed thinking that its main use will in graphical annotations, although one could use it for infographics with multiple copies of each grob, this would go against the grammar of graphics. In this implementation grobs cannot be mapped to an aesthetic through a scale.
As implemented, there is no limitation to the number of insets and faceting is respected. If the base plot shows a map or a bitmap, multiple simple grobs (e.g. national flags) could be superimposed on different countries. The size of the insets is set relative to the main plot, so the combined plot can be scaled.
Please see section Normalised Parent Coordinates below for a description of geom_grob_npc().
geom_vhlines()
This is a convenience geometry that adds both vertical and horizontal guide lines on the same plot layer, using the same syntax as geom_hline() and geom_vline() from package ‘ggplot2’.
ggplot(mpg, aes(displ, hwy, colour = factor(cyl))) +
geom_vhlines(xintercept = c(2.75, 4), yintercept = 27, linetype = "dashed") +
geom_point() +
labs(x = "Engine displacement (l)", y = "Fuel use efficiency (MPG)",
colour = "Engine cylinders\n(number)")
geom_linked_text()
With package ‘ggrepel’ we have become used to easily add text and labels linked to the labeled observations or points by a line segment or arrow. However, sometimes a plot has few text labels and repulsion is unnecessary. In such cases nudging or dodging may be enough, or even preferable so as to achieve consistency.
We can achieve this with geom_linked_text() together with a suitable position function that retains the original x and y coordinates of the point. We currently provide two different, and flexible nudge functions described in a later section. One of them is used by default providing the same default behaviour as geom_text() from ‘ggplot2’ but drawing linking segments if nudging is set to a value different to the default of zero.
Please, be aware that this is a preliminary implementation, whose user interface and behavior are very likely to change in coming months.
my.cars <- mtcars[c(TRUE, FALSE, FALSE, FALSE), ]
my.cars$name <- rownames(my.cars)
my.cars <- my.cars[order(my.cars$wt), ]
ggplot(my.cars, aes(wt, mpg, label = name)) +
geom_point() +
geom_linked_text(aes(colour = factor(cyl)),
angle = 90,
hjust = 1.04, nudge_y = -1,
nudge_x = c(0, 0, -0.05, +0.05, 0, 0 ,0, 0),
arrow = arrow(length = grid::unit(1.5, "mm"))) +
scale_colour_discrete(l = 40) +
expand_limits(y = 0)
Normalised Parent Coordinates
R’s ‘grid’ package defines several units that can be used to describe the locations of plot elements. In ‘ggplot2’ the x and y aesthetics are directly mapped to "native" or data units. For consistent location of annotations with respect to the plotting area we need to rely on "npc" which are expressed relative to the size of the grid viewport. The plotting area in a ggplot is implemented as viewport and support for "npc" coordinates is relatively easy to implement.
To support "npc" positions we have implemented scales for two new aesthetics, npcx and npcy. These are very simple continuous scales which do not support any transformation or changes to their limits, both of which would be meaningless for "npc" units. Variables mapped to these aesthetics can be either numerical with values in the range zero to one or character. A limited set of strings are recognised and converted to "npc" units: "bottom", "center", "top", "left", "middle", "right" ("centre" is a synonym for "center").
To make these scales useful we need also to define geometries that use these new aesthetics. Package ‘ggpmisc’ currently provides geom_text_npc(), geom_label_npc(), geom_table_npc(), geom_plot_npc() and geom_grob_npc().
As is the case for geom_text() and geom_label() from package ‘ggplot2’, "bottom", "center", "top", "left", "middle", "right", plus "inward" and "outward" can be used, as well as numeric values, to control the justification. Justification defaults to "inward" in the geometries described here.
While the usual x and y aesthetics are used whenever the positions of plot elements represent data values, these new scales and geometries are useful only for annotations, i.e., in those cases when we want plot elements at specific positions within the plotting area irrespective of the ranges of the data mapped to the x and y aesthetics. When writing scripts or functions that may be applied to different data sets these new aesthetics help by keeping the code concise and reusable. These geometries are used by default by several of the statistics described in later sections.
As an example let’s imagine that we want to add the structure of a metabolite to a plot. Its position has nothing to do with the data mapped to x and y, so it is conceptually better to use "npc" coordinates. The big practical advantage is that this also allows to keep this part of the plot definition independent of the data being plotted, giving a major advantage in the case of plots with facets with free scale limits. This example can be easily adapted to geom_plot_npc() where a ggplot should mapped to label, and to geom_table_npc() where a data frame should be mapped to label.
We produce the example plot by first constructing a tibble to contain the grob and the coordinate data, and then map these variables to aesthetics using aes(). In the example the tibble has a single row, but this is not a requirement. In this respect these geoms behave as normal geoms, with facets also supported.
file.name <-
system.file("extdata", "Robinin.png",
package = "ggpmisc", mustWork = TRUE)
Robinin <- magick::image_read(file.name)
set.seed(123456)
data.tb <- tibble(x = 1:20, y = (1:20) + rnorm(20, 0, 10))
flavo.tb <- tibble(x = 0.02,
y = 0.95,
width = 1/2,
height = 1/4,
grob = list(grid::rasterGrob(image = Robinin)))
ggplot(data.tb, aes(x, y)) +
geom_grob_npc(data = flavo.tb,
aes(label = grob, npcx = x, npcy = y,
vp.width = width, vp.height = height)) +
geom_point() +
expand_limits(y = 55, x = 0)
Alternatively, we can pass constant values to geom_grob_npc() to obtain the same plot. This approach can be handy in simple cases.
ggplot(data.tb, aes(x, y)) +
geom_grob_npc(label = list(grid::rasterGrob(image = Robinin, width = 1)),
npcx = 0.02, npcy = 0.95,
vp.width = 1/2, vp.height = 1/4) +
geom_point() +
expand_limits(y = 55, x = 0)
We can also use annotate() if the annotation should be the same for all panels, or if we have a single figure panel. In this case there is no need to wrap a single grob in a list.
ggplot(data.tb, aes(x, y)) +
annotate("grob_npc", label = grid::rasterGrob(image = Robinin, width = 1),
npcx = 0.02, npcy = 0.95, vp.width = 1/2, vp.height = 1/4) +
geom_point() +
expand_limits(y = 55, x = 0)
Two additional geometries are based on existing ‘ggplot2’ geometries. They are based on geom_text() and geom_label(). We give an example using geom_text_npc() to produce a “classic” labelling for facets matching the style of theme_classic() and traditional scientific journals’ design.
corner_letters.tb <- tibble(label = LETTERS[1:4],
x = "left",
y = "top",
cyl = c(4,5,6,8))
ggplot(mpg, aes(displ, hwy)) +
geom_point() +
facet_wrap(~cyl, scales = "free") +
geom_text_npc(data = corner_letters.tb,
aes(npcx = x, npcy = y, label = label)) +
theme_classic() +
theme(strip.background = element_blank(),
strip.text.x = element_blank())
Marginal markings
‘ggplot2’ provides geom_rug(), geom_vline() and geom_hline(). Rug plots are intended to be used to represent distributions along the margins of plot. geom_vline() and geom_hline()are normally used to separate regions in a plot or to highlight important values along the x or y axis. When creating plots it is sometimes useful to put small marks along the axes, just inside the plotting area, similar to those in a rug plot, but like geom_vline() and geom_hline() in their purpose.
Three geometries provide such markers: geom_margin_point(), geom_margin_arrow(), and geom_margin_grob(). They behave similarly to geom_vline() and geom_hline() and their positions are determined also by the xintercept and yintercept aesthetics.
In the example below we indicate the group medians along the x axis with filled triangles.
data.tb <- mpg %>%
group_by(cyl) %>%
summarise(hwy = median(hwy), displ = median(displ))
ggplot(mpg, aes(displ, hwy, colour = factor(cyl))) +
geom_x_margin_point(data = data.tb,
aes(xintercept = displ, fill = factor(cyl))) +
expand_limits(y = 10) +
geom_point()
Statistics
stat_centroid()
It can be useful to mark the centroid of a group of observations with a point or with a label. By default stat_centroid() applies function mean() to both x and y by group. If all the values mapped to label within each group are identical, this value is copied to the returned data.
ggplot(mpg, aes(displ, hwy, colour = factor(cyl))) +
geom_point(alpha = 0.33) +
stat_centroid(shape = "cross", size = 4)
Other functions can be passed to this statistic as long as they return a single value that can be mapped to the x and y aesthetics (numeric, time or a factor).
ggplot(mpg, aes(displ, hwy, colour = factor(cyl))) +
geom_point(alpha = 0.33) +
stat_centroid(shape = "cross", size = 4, .fun = median)
The very similar stat_summary_xy() accepts different functions for x and y.
stat_quadrant_counts() and geom_quadrant_lines()
This statistic automates the annotation of plots with number of observations, either by quadrant, by pairs of quadrants or the four quadrants taken together (whole plotting area). Its companion geometry, geom_quadrant_lines() is used in the examples to highlight the quadrants.
We generate some artificial data.
set.seed(4321)
# generate artificial data
x <- -99:100
y <- x + rnorm(length(x), mean = 0, sd = abs(x))
my.data <- data.frame(x,
y,
group = c("A", "B"))
Using defaults except for color.
ggplot(my.data, aes(x, y)) +
geom_quadrant_lines(colour = "red") +
stat_quadrant_counts(colour = "red") +
geom_point() +
expand_limits(y = c(-250, 250))
Pooling quadrants along the x-axis. (pool.along = "y" pools along y.)
ggplot(my.data, aes(x, y)) +
geom_quadrant_lines(colour = "red", pool.along = "x") +
stat_quadrant_counts(colour = "red", pool.along = "x") +
geom_point()
Manual positioning of the text annotations and pooling of all four quadrants, and overriding the default formatting for the label.
ggplot(my.data, aes(x, y)) +
geom_point() +
stat_quadrant_counts(quadrants = 0L, label.x = "left",
aes(label = sprintf("%i observations", stat(count))))
Annotation of only specific quadrants.
ggplot(my.data, aes(x, y)) +
geom_quadrant_lines(colour = "red") +
stat_quadrant_counts(colour = "red", quadrants = c(2, 4)) +
geom_point()
Using facets, even with free scale limits, the labels are placed consistently. This achieved by the default use of geom_text_npc() or as shown below by use of `geom_label_npc(). We expand the y limits to ensure that no observations are occluded by the labels.
ggplot(my.data, aes(x, y, colour = group)) +
geom_quadrant_lines() +
stat_quadrant_counts(geom = "label_npc") +
geom_point() +
expand_limits(y = c(-260, 260)) +
facet_wrap(~group)
stat_apply_group and stat_apply_panel
These statistics apply a function to x and y data. The function is expected to return a vector of the same length as its input. It is usually better to apply such functions through mappings using aes() when they are independent of grouping or to use a transformation for the scale. However, when grouping is important, these statistics solve a problem that would otherwise require pre-computation of the data. Here we plot the difference in circumference between dates for each tree. (As the vector returned by diff() is one element shorter than its input, we prepend NA.) However, scale limits are not recomputed and modifying them would exclude observations. Consequently, in this case we need to manualy set the limit in coord_cartesian().
ggplot(Orange, aes(age, circumference, colour = Tree)) +
stat_apply_group(.fun.y = function(x) {c(NA, diff(x))}, na.rm = TRUE) +
coord_cartesian(xlim = c(420, NA))
stat_peaks() and stat_valleys()
Two statistics make it possible to highlight and/or label peaks and valleys (local maxima and local minima) in a curve. Here we use a time series, using Date objects for time and the default formatting of labels. We also instruct geom_text() not to render overlapping labels. They support numeric, date or datetime mapped to the x aesthetic. Currently a numeric variables has to be mapped to the y aesthetic.
ggplot(lynx, as.numeric = FALSE) + geom_line() +
stat_peaks(colour = "red") +
stat_peaks(geom = "text", colour = "red", vjust = -0.5,
check_overlap = TRUE) +
ylim(-100, 7300)
A plot using numeric values for time and the default formatting of labels. We can also pass other aesthetics recognised the geom_text() such as angle. Here we also highlight and label the valleys.
ggplot(lynx) + geom_line() +
stat_peaks(colour = "red") +
stat_peaks(geom = "text", colour = "red", hjust = -0.2, vjust = 0.5,
angle = 90, check_overlap = TRUE) +
stat_valleys(colour = "blue") +
stat_valleys(geom = "text", colour = "blue", hjust = 1.2, vjust = 0.5,
angle = 90, check_overlap = TRUE) +
expand_limits(y = c(-900, 8000))
We can obtain a very similar plot using Date for time but supplying a format string. In addition marking both peaks and valleys.
ggplot(lynx, as.numeric = FALSE) + geom_line() +
stat_peaks(colour = "red") +
stat_peaks(geom = "text", colour = "red", hjust = -0.2, vjust = 0.5,
angle = 90, check_overlap = TRUE, x.label.fmt = "%Y") +
stat_valleys(colour = "blue") +
stat_valleys(geom = "text", colour = "blue", hjust = 1.2, vjust = 0.5,
angle = 90, check_overlap = TRUE, x.label.fmt = "%Y") +
expand_limits(y = c(-900, 8000))
Using geom_rug for the peaks and valleys.
ggplot(lynx, as.numeric = FALSE) + geom_line() +
stat_peaks(colour = "red") +
stat_peaks(geom = "rug", colour = "red") +
stat_valleys(colour = "blue") +
stat_valleys(geom = "rug", colour = "blue")
stat_dens2d_labels() and stat_dens2d_filter()
These stats had their origin in an enhancement suggestion for ‘ggrepel’ from Hadley Wickham and discussion with Kamil Slowikowski (ggrepel’s author) and others. In fact the code is based on code Kamil gave during the discussion, but simplified and taking a few further ideas from ggplot::stat_dens2d.
Warning! Which observations are selected by the algorithm used, based on MASS:kde2d, depends strongly on the values of parameters h and n. You may need to alter the defaults by passing explicit arguments to these stats. Beware, though, that what are good values, may depend on individual data sets even if they include the same number of observations. For the selection of observations to work cleanly, the argument for n must create a dense grid. Much larger values of n than in the examples in the documentation of MASS::kde2d and ggplot2::stat_dens2d will be needed in most cases.
Some random data with random labels.
random_string <- function(len = 3) {
paste(sample(letters, len, replace = TRUE), collapse = "")
}
# Make random data.
set.seed(1001)
d <- tibble::tibble(
x = rnorm(100),
y = rnorm(100),
group = rep(c("A", "B"), c(50, 50)),
lab = replicate(100, { random_string() })
)
The stat stat_dens2d_filter filters observations, in other words passes to the geom a subset of the data received as input. The default argument for geom is "point".
Using defaults except for the color aesthetic. Highlight 1/4 of observations from lowest density areas of the plot panel.
ggplot(data = d, aes(x, y)) +
geom_point() +
stat_dens2d_filter(keep.fraction = 1/4, colour = "red")
Keep at most 50 observations.
ggplot(data = d, aes(x, y)) +
geom_point() +
stat_dens2d_filter(keep.fraction = 1/4, keep.number = 50, colour = "red")
Keep always 50 observations.
ggplot(data = d, aes(x, y)) +
geom_point() +
stat_dens2d_filter(keep.fraction = 1, keep.number = 50, colour = "red")
Highlighting 1/4 of the observations by under-plotting with larger black points.
ggplot(data = d, aes(x, y, colour = group)) +
stat_dens2d_filter(keep.fraction = 0.25,
size = 3,
colour = "black") +
geom_point()
A different way of highlighting 1/4 of the observations, using over-plotting with a ‘hollow’ shape. We also shift one group with respect to the other.
ggplot(data = d, aes(x + rep(c(-2,2), rep(50,2)),
y, colour = group)) +
geom_point() +
stat_dens2d_filter(shape = 1, size = 3,
keep.fraction = 0.25)
Highlight 1/4 of observations from lowest density areas of the plot, with density considered separately for each individual group. In this case grouping is based on the color aesthetic.
ggplot(data = d, aes(x + rep(c(-2,2), rep(50,2)),
y, colour = group)) +
geom_point() +
stat_dens2d_filter_g(shape = 1, size = 3,
keep.fraction = 0.25)
The stat stat_dens1d_labels replaces the values of the label (aesthetic) variable in data based on density of observations along the x or y axis in the plot panel. The replacement is given by the argument passed to label.fill, which can be a character string or a function accepting a character string as argument and returning also a character string.
The default value for geom is "text". The default value of label.fill is "" which results in empty labels, while using NA as fill label results in observations being omitted. Using NA as label.fill is similar to using stat_dens2d_filter as long as the geom used requires a label aesthetic.
Label 1/10 of observations from lowest density areas of the plot panels.
ggplot(data = d, aes(x, y, label = lab, colour = group)) +
stat_dens2d_labels(keep.fraction = 1/10,
hjust = "outward", vjust = "outward") +
geom_point()
Using the geoms from package ‘ggrepel’ avoids clashes among labels or on top of data points. This works with versions 0.6.0 and newer of ‘ggrepel’. One example with geom_text_repel follows.
ggplot(data = d, aes(x, y, label = lab, colour = group)) +
geom_point() +
stat_dens2d_labels(geom = "text_repel",
keep.fraction = 0.45)
With geom_label_repel one usually needs to use a smaller value for keep.fracton, or a smaller size, as labels use more space on the plot than the test alone.
Additional arguments can be used to change the angle and position of the text, but may give unexpected output when labels are long as the repulsion algorithm “sees” always a rectangular bounding box that is not rotated. With short labels or angles that are multiples of 90 degrees, there is no such problem. Please, see the documentation for ggrepel::geom_text_repel and ggrepel::geom_label_repel for the various ways in which both repulsion and formatting of the labels can be adjusted.
Using NA as argument to label.fill makes the observations with labels set to NA incomplete, and such rows in data are skipped when rendering the plot, before the repulsion algorithm is active. This can lead to overlap between text and points corresponding to unlabelled observations. Whether points are occluded depends on the order of layers and transparency, the occlusion can remain easily unnoticed with geom_label and geom_label_repel. We keep geom_point as the topmost layer to ensure that all observations are visible.
ggplot(data = d, aes(x, y, label = lab, colour = group)) +
stat_dens2d_labels(geom = "label_repel",
keep.fraction = 0.2,
label.fill = NA) +
geom_point()
stat_dens1d_labels() and stat_dens1d_filter()
These stats are similar to stat_dens2d_labels() and stat_dens2d_filter() but compute the density in a single dimension, either the x or y aesthetics.
Warning! Which observations are selected by the algorithm used, based on stats::density, depends strongly on the values of parameters bw, adjust and kernel. You may need to alter the defaults by passing explicit arguments. Beware that what are good values, may depend on individual data sets even if they include the same number of observations. For the selection of observations to work cleanly, the argument for n must large enough to generate a dense grid or the bandwidth may need to be increased by passing a number > 1 as argument. Increasing the bandwidth makes the empirical density function smoother, and the selection of points less dependent on immediate neighbors.
We use the same data as in the previous sections.
random_string <- function(len = 6) {
paste(sample(letters, len, replace = TRUE), collapse = "")
}
# Make random data.
set.seed(1001)
d <- tibble::tibble(
x = rnorm(100),
y = rnorm(100),
group = rep(c("A", "B"), c(50, 50)),
lab = replicate(100, { random_string() })
)
The stat stat_dens1d_filter filters observations, in other words passes to the geom a subset of the data received as input. The default value for geom is "point" and the default orientation is "x".
Using defaults except for the color aesthetic, we highlight 1/4 of observations from lowest density region along the x axis of the plot panel.
ggplot(data = d, aes(x, y)) +
geom_point() +
stat_dens1d_filter(keep.fraction = 0.25,
colour = "red")
We repeat the example above, we highlight 1/4 of observations, but now from lowest density region along the y axis of the plot panel.
ggplot(data = d, aes(x, y)) +
geom_point() +
stat_dens1d_filter(keep.fraction = 0.25,
colour = "red",
orientation = "y")
In other respects than orientation and the parameters passed internally to stats::density() the examples given earlier for stat_dens2d_filter() also apply.
The stat stat_dens1d_labels replaces the values of the label (aesthetic) variable in data based on density of observations along the x or y axis in the plot panel. The replacement is given by the argument passed to label.fill, which can be a character string or a function accepting a character string as argument and returning also a character string.
The default value for geom is "text". The default value of label.fill is "" which results in empty labels, while using NA as fill label results in observations being omitted. Using NA as label.fill is similar to using stat_dens2d_filter as long as the geom used requires a label aesthetic.
Label 1/10 of observations from lowest density regions along x in the plot panel. Normally a repulsive geom is most useful.
ggplot(data = d, aes(x, y, label = lab, colour = group)) +
geom_point() +
stat_dens1d_labels(geom = "text_repel")
Similarly along the y axis.
ggplot(data = d, aes(x, y, label = lab, colour = group)) +
geom_point() +
stat_dens1d_labels(geom = "text_repel", orientation = "y")
In other respects than orientation and the parameters passed internally to stats::density() the examples given earlier for stat_dens2d_labels() also apply.
Position functions
Function position_nudge() from package ‘ggplot2’ simply applies the nudge, or x and y shift based directly on the values passed to its parameters x and y. A new variation on simple nudge is provided by function position_nudge_to().
Two new position functions supporting advanced nudging are made available by package ‘ggpmisc’. While position_nudge() implements shifts along \(x\) and \(y\) axes that are the same for every observation, the new functions compute the direction of the nudging specifically for each observation.
One can think of this approach as nudging away from a focal point, a line or a curve. It is deterministic, while repulsion is aleatory. Advanced nudging works very well, but only for some patterns of observations and may require manual adjustment of positions, repulsion is more generally applicable.
All the new position functions save the original coordinates to allow linking the nudged objects, usually text labels, to be linked to the original position. This is currently only supported by geom_linked_text(), but they may be also supported by ‘ggrepel’ in the future.
position_nudge_keep()
position_nudge_keep() (and position_nudge_center() With the default direction = "none") behaves as ggplot2::position_nudge() but saves the original coordinates. Currently the saved original coordinates are used only by geom_linked_text() as shown in the corresponding section above.
set.seed(84532)
df <- data.frame(
x = rnorm(10),
y = rnorm(10, 2, 2),
l = letters[1:10]
)
Here we keep the default justification, with text centred on its location.
ggplot(df, aes(x, y, label = l)) +
geom_point() +
geom_text(position = position_nudge_keep(x = 0.1, y = 0.15))
When using geom_linked_text() we need to change the default justification.
ggplot(df, aes(x, y, label = l)) +
geom_point() +
geom_linked_text(position = position_nudge_keep(x = 0.1, y = 0.15),
hjust = -0.8, vjust = 0.2)
As this is also the default position function of this geometry, the code below is equivalent.
ggplot(df, aes(x, y, label = l)) +
geom_point() +
geom_linked_text(nudge_x = 0.1, nudge_y = 0.15, hjust = -0.8, vjust = 0.2)
position_nudge_to()
Function position_nudge_to() differs from ggplot2::position_nudge() in that the coordinates of the new positions are supplied directly rather than as a displacement.
set.seed(84532)
df <- data.frame(
x = rnorm(8),
y = rnorm(8, 2, 2),
l = letters[1:8]
)
We here the pass a vector of length one as argument for y, but x and y also accept vectors of the same length as rows has data. In other words, it allows manual positioning of text and labels.
ggplot(df, aes(x, y, label = l)) +
geom_point() +
geom_text(position = position_nudge_to(y = 6))
position_nudge_center()
Function position_nudge_center() can nudge radially away from a focal point if both x and y are passed as arguments, or towards opposite sides of a boundary line if only one of x or y is passed an argument. By default, the “center” is the centroid computed using mean(), but other functions or numeric values can be passed to override it.
set.seed(84532)
df <- data.frame(
x = rnorm(20),
y = rnorm(20, 2, 2),
l = sample(letters[1:6], 20, replace = TRUE)
)
When using one of the directions "split" or "radial" by default a centroid computed with mean() is used as focal point. If the direction is "split" then only the sign of x and/or y are reversed to the left and below the focal point, respectively.
ggplot(df, aes(x, y, label = l)) +
geom_point() +
geom_text(position = position_nudge_center(x = 0.1,
y = 0.15,
direction = "split"))
By default x and y take a value of zero.
ggplot(df, aes(x, y, label = l)) +
geom_point() +
geom_text(position = position_nudge_center(x = 0.1,
direction = "split"))
We can override the default center coordinates by passing numeric values to center_x and/or center_y.
ggplot(df, aes(x, y, label = l)) +
geom_point() +
geom_text(position = position_nudge_center(x = 0.1,
center_x = 1,
direction = "split"))
We can override the default center coordinates computed with mean() by passing as argument a different function.
ggplot(df, aes(x, y, label = l)) +
geom_point() +
geom_text(position = position_nudge_center(x = 0.1,
center_x = median,
direction = "split"))
Direction "radial" applies nudge radially away from the focal point.
ggplot(df, aes(x, y, label = l)) +
geom_point() +
geom_text(position = position_nudge_center(x = 0.1,
y = 0.3,
direction = "radial"))
position_nudge_line()
Function position_nudge_line() nudges away from a line, which can be a user supplied straight line as well as a smooth spline or a polynomial fitted to the observations themselves. The nudging is away and perpendicular to the local slope of the straight or curved line. It relies on the same assumptions as linear regression, assuming that x values are not subject to error. This in most cases prevents labels from overlaping a curve fitted to the data, even if not exactly based on the same model fit. When observations are sparse, this may be enough to obtain a nice arrangement of data labels.
set.seed(16532)
df <- data.frame(
x = -10:10,
y = (-10:10)^2,
yy = (-10:10)^2 + rnorm(21, 0, 4),
yyy = (-10:10) + rnorm(21, 0, 4),
l = letters[1:21]
)
The first, simple example shows that position_nudge_line() has shifted the direction of the nudging based on the alignment of the observations along a line. One could, of course, have in this case passed suitable values as arguments to x and y using position_nudge() from package ‘ggplot2’.
Function position_nudge_line() will work without change irrespective of the slope or intercept along which the observations fall.
ggplot(df, aes(x, -3 * x, label = l)) +
geom_point() +
geom_abline(intercept = 0, slope = -3, linetype = "dotted") +
geom_text(position = position_nudge_line(x = -0.5, y = -0.8))
With observations with high variation in y, a linear model fit may need to be used.
ggplot(subset(df, x >= 0), aes(x, yyy)) +
geom_point() +
stat_smooth(method = "lm", formula = y ~ x) +
geom_text(aes(label = l),
vjust = "center", hjust = "center",
position = position_nudge_line(x = 0, y = 1.2,
method = "lm", formula = y ~ x,
direction = "split"))
With lower variation in y, we can pass to line_nudge a multiplier to keep labels outside of the confidence band.
ggplot(subset(df, x >= 0), aes(y, yy)) +
geom_point() +
stat_smooth(method = "lm", formula = y ~ x) +
geom_text(aes(label = l),
position = position_nudge_line(method = "lm",
x = 3, y = 3,
line_nudge = 2.8,
direction = "split"))
If we want the nudging based on an arbitrary straight line not computed from data, we can pass the intercept and slope in a numeric vector of length two as an argument to parameter abline.
ggplot(subset(df, x >= 0), aes(y, yy)) +
geom_point() +
geom_abline(intercept = 0, slope = 1, linetype = "dotted") +
geom_text(aes(label = l),
position = position_nudge_line(abline = c(0, 1),
x = 3, y = 3,
direction = "split"))
With observations that follow exactly a simple curve the defaults work well to automate the nudging of individual data labels away from the implicit curve. Positive values as arguments to x and y correspond to above and inside the curve. One could, of course, pass also in this case suitable values as arguments to x and y using position_nudge() from package ‘ggplot2’, but these arguments would need to be vectors with different nudge values for each observation.
ggplot(df, aes(x, y, label = l)) +
geom_point() +
geom_line(linetype = "dotted") +
geom_text(position = position_nudge_line(x = 0.6, y = 6))
Negative values passed as arguments to x and y correspond to labels below and outside the curve.
ggplot(df, aes(x, y, label = l)) +
geom_point() +
geom_line(linetype = "dotted") +
geom_text(position = position_nudge_line(x = -0.6, y = -6))
When the observations include random variation along y, it is important that the smoother used for the line added to a plot and that passed to position_nudge_line() are similar. By default stat_smooth() uses "loess" and position_nudge_line() with method "spline", smooth.sline(), which are a good enough match.
ggplot(df, aes(x, yy, label = l)) +
geom_point() +
stat_smooth(formula = y ~ x, method = "loess") +
geom_text(aes(y = yy, label = l),
position = position_nudge_line(x = 0.6,
y = 6,
direction = "split"))
When fitting a polynomial, "lm" should be the argument passed to method and a model formula preferably based on poly(), setting raw = TRUE, as argument to formula. Currently no other methods are implemented in position_nudge_line().
ggplot(df, aes(x, yy, label = l)) +
geom_point() +
stat_smooth(method = "lm", formula = y ~ poly(x, 2, raw = TRUE)) +
geom_text(aes(y = yy, label = l),
position = position_nudge_line(method = "lm",
x = 0.6,
y = 6,
formula = y ~ poly(x, 2, raw = TRUE),
direction = "split"))
Grouping is supported.
df <- data.frame(x = rep(1:10, 2),
y = c(1:10, 10:1),
group = rep(c("a", "b"), c(10, 10)),
l = "+")
ggplot(df, aes(x, y, label = l, color = group)) +
geom_line(linetype = "dotted") +
geom_point() +
geom_text(position = position_nudge_line(x = 0.25, y = 1)) +
coord_equal(ratio = 0.5)
One needs to ensure that grouping is in effect in the geoms with nudging.
ggplot(df, aes(x, y, label = l, color = group, group = group)) +
geom_line(linetype = "dotted") +
geom_text() +
geom_text(color = "red",
position = position_nudge_line(x = 1, y = 1)) +
geom_text(color = "blue",
position = position_nudge_line(x = -1, y = -1)) +
coord_equal()
Facets are also supported.
ggplot(df, aes(x, y, label = l)) +
geom_line(linetype = "dotted") +
geom_point() +
geom_text(position = position_nudge_line(x = 1, y = 1),
color = "red") +
geom_text(position = position_nudge_line(x = -1, y = -1),
color = "blue") +
facet_wrap(~group) +
coord_equal(ratio = 1)
Appendix A: More on density filtering
We define a function to simplify the generation of random data sets based on different probability distributions.
make_data_tbl <- function(nrow = 100, rfun = rnorm, ...) {
if (nrow %% 2) {
nrow <- nrow + 1
}
set.seed(1001)
tibble::tibble(
x = rfun(nrow, ...),
y = rfun(nrow, ...),
group = rep(c("A", "B"), c(nrow / 2, nrow / 2))
)
}
In all the examples in this vignette we use colors to demonstrate which data points are selected, but any other suitable aesthetic and discrete scale can be used instead. With keep.sparse = FALSE we keep 1/3 of the observations in the denser region of the plot. Although here we first plot all data points and later overplot the selected ones this is not necessary.
ggplot(data = make_data_tbl(300), aes(x, y)) +
geom_point() +
stat_dens2d_filter(colour = "red",
keep.sparse = FALSE,
keep.fraction = 1/3)
Here we highlight the observations split in three group equal groups, each of a different density of observations.
ggplot(data = make_data_tbl(300), aes(x, y)) +
geom_point() +
stat_dens2d_filter(colour = "red",
keep.sparse = FALSE,
keep.fraction = 1/3)+
stat_dens2d_filter(colour = "blue",
keep.fraction = 1/3)
The algorithm seems to work well also with other distributions, in this example the uniform distribution.
ggplot(data = make_data_tbl(300, rfun = runif), aes(x, y)) +
geom_point() +
stat_dens2d_filter(colour = "red", keep.fraction = 1/2)
One example with the gamma distribution, which is asymmetric.
ggplot(data = make_data_tbl(300, rfun = rgamma, shape = 2),
aes(x, y)) +
geom_point() +
stat_dens2d_filter(colour = "red", keep.fraction = 1/3)
