Cookbook

Every exported function and every option that changes the output, as worked examples. Each figure below was rendered by ggnext when this page was built, and several sections check the computed numbers against base R.

This document exercises every exported function of ggnext and every option that changes what is drawn. It doubles as a regression check: if any layer, scale, coordinate system or theme setting breaks, this document fails to knit.

Writing a plot object in a chunk renders it inline — ggnext registers a knit_print() method, so no results='asis' boilerplate is needed.

packageVersion("ggnext")

#> [1] '0.1.0'

length(getNamespaceExports("ggnext"))

#> [1] 154

1. The grammar

1.1 Minimal plot

A plot is data, an aesthetic mapping, and at least one layer.

ggnext(cars, aes(speed, dist)) + geom_point()

1.2 Plots are immutable values

+ returns a new plot, so a partial specification is a reusable template.

base <- ggnext(iris, aes(Sepal.Length, Sepal.Width))
length(base@layers)            # still empty

#> [1] 0

length((base + geom_point())@layers)

#> [1] 1

length(base@layers)            # base is unchanged

#> [1] 0

1.3 aes() — mapping vs setting

aes(speed, dist, color = gear)

#> <ggnext aesthetic mapping>
#>   x -> speed
#>   y -> dist
#>   color -> gear

A constant inside aes() that names a real colour is honoured literally rather than being treated as a one-level category.

ggnext(cars, aes(speed, dist, color = "steelblue")) + geom_point(size = 4)

1.4 Layer-level data and mapping

A layer can override both, which is how you annotate one plot with a second dataset.

means <- aggregate(Sepal.Length ~ Species, iris, mean)

ggnext(iris, aes(Species, Sepal.Length)) +
  geom_jitter(alpha = 0.3) +
  geom_point(aes(Species, Sepal.Length), data = means,
             color = "#C1462F", size = 7)

1.5 Output size

p_wide <- ggnext(cars, aes(speed, dist), width = 900, height = 260) +
  geom_point()
p_wide

plot_size() does the same thing after the fact.

b <- ggnext:::build_geometry(ggnext(cars, aes(speed, dist)) +
                                geom_point() + plot_size(800, 600))
c(width = b$width, height = b$height)

#>  width height 
#>    800    600

2. Essential geoms

2.1 geom_point

ggnext(cars, aes(speed, dist)) + geom_point(color = "#2B6BE0", size = 4,
                                             alpha = 0.7)

Size and colour can be mapped instead of set:

ggnext(iris, aes(Sepal.Length, Sepal.Width,
                  color = Species, size = Petal.Length)) +
  geom_point(alpha = 0.75)

2.2 geom_jitter

ggnext(iris, aes(Species, Sepal.Width, color = Species)) +
  geom_jitter(alpha = 0.7) +
  theme(legend_position = "none")

Jitter is seeded, so the same plot renders identically every time — and it restores the caller’s random stream:

set.seed(42); before <- .Random.seed
invisible(render(ggnext(iris, aes(Species, Sepal.Width)) + geom_jitter()))
identical(.Random.seed, before)

#> [1] TRUE

2.3 geom_line, geom_path, geom_step

ts <- data.frame(
  t = rep(1:12, 2),
  v = c(cumsum(rnorm(12, 2)), cumsum(rnorm(12, 1))),
  g = rep(c("A", "B"), each = 12)
)
ggnext(ts, aes(t, v, color = g)) + geom_line(linewidth = 2) + theme_minimal()

geom_path() follows data order rather than x order; geom_step() holds each value until the next observation.

steps <- data.frame(t = 1:8, v = c(2, 2, 3, 3, 5, 4, 4, 6))
ggnext(steps, aes(t, v)) + geom_step(color = "#12A594", linewidth = 2) +
  geom_point() + theme_minimal()

Dashed lines via dash (an SVG dash pattern):

ggnext(steps, aes(t, v)) + geom_line(dash = "6,4") + theme_minimal()

2.4 geom_area and geom_ribbon

ggnext(data.frame(t = 1:10, v = c(2, 4, 3, 6, 8, 7, 9, 8, 11, 13)),
        aes(t, v)) +
  geom_area(alpha = 0.45) + theme_minimal()

band <- data.frame(t = 1:12, lo = (1:12) * 0.8, hi = (1:12) * 1.4)
ggnext(band, aes(t, ymin = lo, ymax = hi)) +
  geom_ribbon(alpha = 0.4) + theme_minimal()

2.5 geom_segment and geom_dumbbell

seg <- data.frame(x = 1:4, y = c(2, 4, 3, 5),
                  xe = (1:4) + 0.7, ye = c(4, 6, 2, 7))
ggnext(seg, aes(x, y, xend = xe, yend = ye)) +
  geom_segment(linewidth = 2) + theme_minimal()

db <- data.frame(g = c("North", "South", "East", "West"),
                 before = c(12, 18, 9, 14), after = c(19, 21, 15, 13))
ggnext(db, aes(before, xend = after, y = g)) +
  geom_dumbbell() + theme_minimal()

2.6 Reference lines

geom_hline() and geom_vline() span the panel and ignore the plot’s aes(), so they never disturb the data mapping.

ggnext(cars, aes(speed, dist)) +
  geom_point(alpha = 0.6) +
  geom_hline(mean(cars$dist), dash = "5,4", color = "#C1462F") +
  geom_vline(mean(cars$speed), dash = "5,4", color = "#C1462F") +
  theme_minimal()

2.7 geom_text

lab <- data.frame(x = c(1, 2, 3), y = c(3, 1, 2),
                  l = c("alpha", "beta", "gamma"))
ggnext(lab, aes(x, y, label = l)) +
  geom_point(size = 6, alpha = 0.25) + geom_text() + theme_minimal()

2.8 geom_tile

grid <- expand.grid(x = 1:10, y = 1:8)
grid$z <- as.vector(outer(1:10, 1:8, function(a, b) sin(a / 2) + cos(b / 2)))
ggnext(grid, aes(x, y, color = z)) + geom_tile() + theme_minimal()

3. Distributions

3.1 geom_bar and geom_col

geom_bar() counts rows; geom_col() takes the height from y.

cnt <- data.frame(g = c("a", "a", "a", "b", "b", "c"))
ggnext(cnt, aes(g)) + geom_bar() + theme_minimal()

vals <- data.frame(g = c("alpha", "beta", "gamma", "delta"),
                   v = c(12, 27, 19, 8))
ggnext(vals, aes(g, v, color = g)) + geom_col() +
  theme(legend_position = "none")

3.2 Position adjustments

grp <- data.frame(g = rep(c("Q1", "Q2", "Q3"), each = 3),
                  grp = rep(c("a", "b", "c"), 3),
                  v = c(4, 6, 3, 7, 4, 5, 5, 8, 2))
ggnext(grp, aes(g, v, color = grp)) + geom_col(position = "stack") +
  labs(title = 'position = "stack" (default)') + theme_minimal()

ggnext(grp, aes(g, v, color = grp)) + geom_col(position = "dodge") +
  labs(title = 'position = "dodge"') + theme_minimal()

ggnext(grp, aes(g, v, color = grp)) + geom_col(position = "fill") +
  labs(title = 'position = "fill" (proportions)') + theme_minimal()

Stacked totals train the axis correctly — the domain covers the stack, not the tallest single bar:

stacked <- ggnext:::build_geometry(
  ggnext(grp, aes(g, v, color = grp)) + geom_col(position = "stack")
)
stacked$panels[[1]]$y$domain

#> [1] -0.8 16.8

3.3 geom_histogram

bins = n yields exactly n bins, and counts sum to the row count.

ggnext(cars, aes(speed)) + geom_histogram(bins = 8) + theme_minimal()

h <- plot_data(ggnext(cars, aes(speed)) + geom_histogram(bins = 5))
h

#>      x  y ymin ymax xmin xmax
#> 1  6.1  5    0    5  4.0  8.2
#> 2 10.3 10    0   10  8.2 12.4
#> 3 14.5 13    0   13 12.4 16.6
#> 4 18.7 15    0   15 16.6 20.8
#> 5 22.9  7    0    7 20.8 25.0

c(bins = nrow(h), total = sum(h$y), rows = nrow(cars))

#>  bins total  rows 
#>     5    50    50

binwidth overrides bins:

ggnext(cars, aes(speed)) + geom_histogram(binwidth = 5) + theme_minimal()

3.4 geom_density

ggnext(iris, aes(Sepal.Length, color = Species)) +
  geom_density(alpha = 0.45) + theme_minimal()

The estimate integrates to 1:

dd <- compute_stat(stat_density(), list(x = rnorm(500),
                                        group = rep("all", 500)))
round(sum(dd$y) * diff(dd$x[1:2]), 3)

#> [1] 1

3.5 geom_boxplot

ggnext(iris, aes(Species, Sepal.Length, color = Species)) +
  geom_boxplot() + theme(legend_position = "none")

Quartiles match stats::quantile():

bx <- plot_data(ggnext(iris, aes(Species, Sepal.Length)) + geom_boxplot())
setosa <- iris$Sepal.Length[iris$Species == "setosa"]
c(computed = bx$middle[1], base_r = median(setosa))

#> computed   base_r 
#>        5        5

3.6 geom_violin

ggnext(iris, aes(Species, Sepal.Width, color = Species)) +
  geom_violin() + theme(legend_position = "none")

3.7 Layered distribution view

Layers draw in the order added: shape underneath, summary, raw data on top.

ggnext(iris, aes(Species, Sepal.Length, color = Species)) +
  geom_violin(alpha = 0.25) +
  geom_boxplot() +
  geom_jitter(alpha = 0.4) +
  theme(legend_position = "none")

3.8 geom_ridgeline

ggnext(iris, aes(Sepal.Length, Species)) +
  geom_ridgeline() + theme_minimal()

ggnext(iris, aes(Sepal.Length, Species)) +
  geom_ridgeline(scale = 3, alpha = 0.55) +
  labs(title = "scale = 3 makes ridges overlap") + theme_minimal()

3.9 geom_smooth

ggnext(cars, aes(speed, dist)) +
  geom_point(alpha = 0.6) + geom_smooth(method = "lm") + theme_minimal()

The linear fit matches stats::lm():

sm <- compute_stat(stat_smooth(method = "lm"),
                   list(x = cars$speed, y = cars$dist,
                        group = rep("all", nrow(cars))))
fit <- lm(dist ~ speed, cars)
max(abs(sm$y - unname(predict(fit, data.frame(speed = sm$x)))))

#> [1] 0

ggnext(cars, aes(speed, dist)) +
  geom_point(alpha = 0.6) + geom_smooth(method = "loess", se = FALSE) +
  theme_minimal()

3.10 Error bars and point ranges

eb <- data.frame(g = c("A", "B", "C", "D"), m = c(5, 7, 4, 8),
                 lo = c(4, 6.2, 3.1, 7.1), hi = c(6, 7.8, 4.9, 8.9))
ggnext(eb, aes(g, m, ymin = lo, ymax = hi)) +
  geom_errorbar() + theme_minimal()

ggnext(eb, aes(g, m, ymin = lo, ymax = hi)) +
  geom_pointrange() + theme_minimal()

3.11 geom_waterfall

wf <- data.frame(
  step = factor(c("Start", "Sales", "Costs", "Tax", "End"),
                levels = c("Start", "Sales", "Costs", "Tax", "End")),
  v = c(100, 45, -30, -12, 0)
)
ggnext(wf, aes(step, v)) + geom_waterfall() + theme_minimal()

4. Scales and axes

4.1 Axis titles, breaks, labels, padding

ggnext(cars, aes(speed, dist)) +
  geom_point() +
  scale_x_continuous(
    name = "Speed (mph)",
    breaks = c(5, 10, 15, 20, 25),
    expand = 0
  ) +
  scale_y_continuous(
    name = "Stopping distance",
    labels = function(v) paste0(v, " ft")
  ) +
  theme_minimal()

expand = 0 makes the domain exactly the data range:

tight <- ggnext:::build_geometry(
  ggnext(cars, aes(speed, dist)) + geom_point() +
    scale_x_continuous(expand = 0)
)
rbind(domain = tight$panels[[1]]$x$domain, data = range(cars$speed))

#>        [,1] [,2]
#> domain    4   25
#> data      4   25

4.2 Limits

ggnext(cars, aes(speed, dist)) + geom_point() +
  xlim(0, 30) + ylim(0, 150) + theme_minimal()

lims() sets both at once. Limits set the domain, not a crop — data outside still goes through the stats and is clipped when drawn.

ggnext(cars, aes(speed, dist)) + geom_point() +
  lims(x = c(10, 20)) +
  geom_smooth(method = "lm") +
  labs(subtitle = "the fit still uses all 50 rows") + theme_minimal()

4.3 Transforms

lg <- data.frame(x = 10^(1:5), y = 1:5)
ggnext(lg, aes(x, y)) + geom_point(size = 5) + scale_x_log10() +
  theme_minimal()

Positions are the log of the data; labels stay in original units.

lb <- ggnext:::build_geometry(
  ggnext(lg, aes(x, y)) + geom_point() + scale_x_log10()
)
unlist(lb$panels[[1]]$x$ticks$labels)

#> [1] "10"     "100"    "1000"   "10000"  "100000"

sq <- data.frame(x = c(1, 4, 9, 16, 25), y = 1:5)
ggnext(sq, aes(x, y)) + geom_point(size = 5) + scale_x_sqrt() +
  theme_minimal()

ggnext(sq, aes(x, y)) + geom_point(size = 5) + scale_y_reverse() +
  labs(title = "scale_y_reverse()") + theme_minimal()

4.4 Discrete scales

Explicit level order — how you sort bars by size rather than alphabetically.

srt <- data.frame(name = c("delta", "alpha", "gamma", "beta"),
                  value = c(8, 12, 19, 27))
srt <- srt[order(-srt$value), ]
ggnext(srt, aes(name, value)) + geom_col() +
  scale_x_discrete(limits = srt$name) + theme_minimal()

4.5 Colour scales

ggnext(iris, aes(Sepal.Length, Sepal.Width, color = Species)) +
  geom_point(size = 3) +
  scale_color_manual(c("#2B6BE0", "#E05A2B", "#12A594")) +
  theme_minimal()

ggnext(iris, aes(Sepal.Length, Sepal.Width, color = Petal.Length)) +
  geom_point(size = 3) +
  scale_color_gradient(low = "#FFF3B0", high = "#9E2A2B") +
  labs(color = "Petal length") + theme_minimal()

5. Coordinates

5.1 coord_flip

ggnext(vals, aes(g, v, color = g)) + geom_col() + coord_flip() +
  theme(legend_position = "none")

5.2 coord_polar

days <- data.frame(d = c("Mon", "Tue", "Wed", "Thu", "Fri", "Sat", "Sun"),
                   v = c(4, 7, 6, 9, 12, 15, 11))
ggnext(days, aes(d, v, color = d)) + geom_col() + coord_polar() +
  theme(legend_position = "none")

Donut hole, direction and start angle:

ggnext(days, aes(d, v, color = d)) + geom_col() +
  coord_polar(inner = 0.35, direction = -1, start = pi / 4) +
  theme(legend_position = "none")

Categories wrap evenly around the full turn:

pol <- ggnext:::build_geometry(
  ggnext(days, aes(d, v)) + geom_col() + coord_polar()
)
unlist(pol$panels[[1]]$polar$spokes)

#> [1] 0.0000000 0.1428571 0.2857143 0.4285714 0.5714286 0.7142857 0.8571429

6. Facets

6.1 facet_wrap

ggnext(iris, aes(Sepal.Length, Sepal.Width, color = Species)) +
  geom_point() + facet_wrap(Species) +
  theme(legend_position = "none")

6.2 Grid shape and free scales

ggnext(iris, aes(Sepal.Length, Sepal.Width, color = Species)) +
  geom_point() + facet_wrap(Species, ncol = 2, scales = "free") +
  theme(legend_position = "none")

Fixed scales share one domain; free scales train per panel.

fixed <- ggnext:::build_geometry(
  ggnext(iris, aes(Sepal.Length, Sepal.Width)) + geom_point() +
    facet_wrap(Species))
free <- ggnext:::build_geometry(
  ggnext(iris, aes(Sepal.Length, Sepal.Width)) + geom_point() +
    facet_wrap(Species, scales = "free"))
rbind(
  fixed_p1 = fixed$panels[[1]]$x$domain,
  fixed_p3 = fixed$panels[[3]]$x$domain,
  free_p1  = free$panels[[1]]$x$domain,
  free_p3  = free$panels[[3]]$x$domain
)

#>           [,1]  [,2]
#> fixed_p1 4.120 8.080
#> fixed_p3 4.120 8.080
#> free_p1  4.225 5.875
#> free_p3  4.750 8.050

6.3 facet_grid

mt <- transform(mtcars, cyl = factor(cyl), am = factor(am))
ggnext(mt, aes(disp, mpg)) + geom_point() + facet_grid(am, cyl) +
  theme_minimal()

6.4 A layer without the faceting variable repeats in every panel

ggnext(iris, aes(Sepal.Length, Sepal.Width)) +
  geom_point(alpha = 0.6) +
  geom_hline(3, color = "#C1462F", dash = "4,3") +
  facet_wrap(Species) + theme_minimal()

7. Titles, labels and themes

7.1 The title block

ggnext(cars, aes(speed, dist)) +
  geom_point() +
  labs(
    title    = "Stopping distance rises with speed",
    subtitle = "1920s road tests, 50 observations",
    caption  = "Source: datasets::cars",
    tag      = "A",
    x = "Speed (mph)", y = "Distance (ft)"
  ) +
  theme_minimal()

labs() merges across repeated calls; ggtitle(), xlab(), ylab() are shorthands.

lab_plot <- ggnext(cars, aes(speed, dist)) + geom_point() +
  labs(title = "first") + labs(subtitle = "second") + xlab("X") + ylab("Y")
b <- ggnext:::build_geometry(lab_plot)
unlist(b$labels[c("title", "subtitle")])

#>    title subtitle 
#>  "first" "second"

7.2 The six presets

theme_demo <- function(th, nm) {
  ggnext(iris, aes(Sepal.Length, Sepal.Width, color = Species)) +
    geom_point(alpha = 0.85) + labs(title = nm) + th
}
theme_demo(theme_ggnext(), "theme_ggnext()")

theme_demo(theme_minimal(), "theme_minimal()")

theme_demo(theme_classic(), "theme_classic()")

theme_demo(theme_modern(), "theme_modern()")

theme_demo(theme_dark(), "theme_dark()")

theme_demo(theme_void(), "theme_void()")

7.3 Customising a theme

base picks the preset to start from; everything else applies over it.

ggnext(iris, aes(Sepal.Length, Sepal.Width, color = Species)) +
  geom_point(size = 3) +
  labs(title = "Custom theme") +
  theme(
    base            = theme_minimal(),
    panel_fill      = "#FFF8F0",
    grid_major_x    = FALSE,
    grid_color      = "#E0D5C8",
    plot_title_size = 22,
    legend_position = "bottom",
    point_palette   = c("#2B6BE0", "#E05A2B", "#12A594")
  )

7.4 All 35 theme settings

defaults <- ggnext:::THEME_DEFAULTS
data.frame(
  setting = names(defaults),
  default = vapply(defaults, function(v) {
    if (length(v) == 0) "—" else paste(format(v), collapse = ", ")
  }, character(1)),
  row.names = NULL
)

#>               setting                      default
#> 1                name                       ggnext
#> 2          background                      #FFFFFF
#> 3          panel_fill                      #F4F4F6
#> 4          grid_color                      #FFFFFF
#> 5    grid_color_minor                             
#> 6          axis_color                      #3A3A3A
#> 7         label_color                      #3A3A3A
#> 8         title_color                      #1A1A22
#> 9      subtitle_color                      #5A5A66
#> 10         strip_fill                      #E4E4EA
#> 11        strip_color                      #2A2A33
#> 12  legend_text_color                      #3A3A3A
#> 13       panel_border                             
#> 14               font Helvetica, Arial, sans-serif
#> 15         title_font                             
#> 16     tick_font_size                           11
#> 17    title_font_size                           13
#> 18    plot_title_size                           17
#> 19 plot_subtitle_size                           12
#> 20       caption_size                           10
#> 21    strip_font_size                           11
#> 22   legend_font_size                           11
#> 23         title_face                         bold
#> 24           tick_len                            5
#> 25       grid_major_x                         TRUE
#> 26       grid_major_y                         TRUE
#> 27        axis_line_x                         TRUE
#> 28        axis_line_y                         TRUE
#> 29            ticks_x                         TRUE
#> 30            ticks_y                         TRUE
#> 31        axis_text_x                         TRUE
#> 32        axis_text_y                         TRUE
#> 33       axis_title_x                         TRUE
#> 34       axis_title_y                         TRUE
#> 35    legend_position                        right
#> 36      point_palette                            —
#> 37       gradient_low                             
#> 38      gradient_high

Individual toggles, verified against the rendered SVG:

count_tag <- function(p, tag) {
  svg <- render(p)
  length(gregexpr(tag, svg, fixed = TRUE)[[1]]) *
    (!grepl("^\\s*$", svg)) * as.integer(grepl(tag, svg, fixed = TRUE))
}
p0 <- ggnext(cars, aes(speed, dist)) + geom_point()
data.frame(
  setting = c("grid on (default)", "grid_major_x = FALSE",
              "grid_color = ''", "axis_text_x = FALSE"),
  gridlines_or_labels = c(
    count_tag(p0, "stroke=\"#FFFFFF\""),
    count_tag(p0 + theme(grid_major_x = FALSE), "stroke=\"#FFFFFF\""),
    count_tag(p0 + theme(grid_color = ""), "stroke=\"#FFFFFF\""),
    count_tag(p0 + theme(axis_text_x = FALSE), "text-anchor=\"middle\"")
  )
)

#>                setting gridlines_or_labels
#> 1    grid on (default)                  12
#> 2 grid_major_x = FALSE                   7
#> 3      grid_color = ''                   0
#> 4  axis_text_x = FALSE                   2

7.5 Legends

ggnext(iris, aes(Sepal.Length, Sepal.Width, color = Species)) +
  geom_point() + theme(legend_position = "bottom")

ggnext(iris, aes(Sepal.Length, Sepal.Width, color = Species)) +
  geom_point() + theme(legend_position = "none")

8. Layout geoms

8.1 geom_radar

radar <- data.frame(
  axis  = rep(c("Speed", "Power", "Range", "Cost", "Safety"), 2),
  value = c(8, 6, 7, 4, 9, 5, 9, 4, 8, 6),
  model = rep(c("A", "B"), each = 5)
)
ggnext(radar, aes(axis, value, color = model)) +
  geom_radar() + coord_polar() + theme_minimal()

8.2 geom_treemap

tm <- data.frame(region = c("North", "South", "East", "West", "Central"),
                 revenue = c(52, 38, 27, 19, 11))
ggnext(tm, aes(size = revenue, label = region, color = region)) +
  geom_treemap() + theme(legend_position = "none")

8.3 geom_sankey

flows <- data.frame(
  from = c("Visited", "Visited", "Signed up", "Signed up"),
  to   = c("Signed up", "Left", "Purchased", "Churned"),
  n    = c(400, 600, 150, 250)
)
ggnext(flows, aes(x = from, xend = to, y = n)) + geom_sankey()

8.4 geom_network

net <- data.frame(from = c("A", "A", "B", "C", "D", "E", "B"),
                  to   = c("B", "C", "C", "D", "E", "A", "E"))
ggnext(net, aes(x = from, xend = to)) + geom_network()

8.5 geom_chord

ch <- data.frame(from = c("A", "A", "B", "C"), to = c("B", "C", "C", "A"),
                 n = c(5, 3, 7, 2))
ggnext(ch, aes(x = from, xend = to, y = n)) + geom_chord()

8.6 geom_stream

st <- data.frame(
  t = rep(1:8, 3),
  v = c(2, 4, 6, 5, 3, 2, 4, 5, 1, 3, 5, 8, 6, 4, 3, 2, 5, 4, 3, 4, 6, 7, 5, 4),
  grp = rep(c("a", "b", "c"), each = 8)
)
ggnext(st, aes(t, v, color = grp)) + geom_stream() + theme_minimal()

8.7 geom_bump

bump <- data.frame(
  year = rep(2020:2024, 4),
  rank = c(1, 2, 3, 3, 2, 2, 1, 1, 2, 1, 3, 3, 2, 1, 3, 4, 4, 4, 4, 4),
  team = rep(c("A", "B", "C", "D"), each = 5)
)
ggnext(bump, aes(year, rank, color = team)) +
  geom_bump() + scale_y_reverse() + theme_minimal()

8.8 geom_funnel

fn <- data.frame(
  stage = factor(c("Visits", "Signups", "Trials", "Paid"),
                 levels = c("Visits", "Signups", "Trials", "Paid")),
  n = c(10000, 3200, 1100, 420)
)
ggnext(fn, aes(stage, n, color = stage)) + geom_funnel() +
  theme(legend_position = "none")

8.9 geom_parallel

sub <- iris[c(1, 20, 60, 80, 110, 140), ]
par_d <- data.frame(
  id  = rep(rownames(sub), 4),
  var = rep(c("SL", "SW", "PL", "PW"), each = nrow(sub)),
  val = c(sub$Sepal.Length, sub$Sepal.Width, sub$Petal.Length, sub$Petal.Width),
  sp  = rep(as.character(sub$Species), 4)
)
ggnext(par_d, aes(var, val, group = id, color = sp)) +
  geom_parallel() + theme_minimal()

8.10 geom_upset

us <- data.frame(sets = c("A", "A&B", "B", "A&B&C", "C", "A&B",
                          "A", "B&C", "A&C", "A&B"))
ggnext(us, aes(label = sets)) + geom_upset()

9. Machine-learning geoms

9.1 geom_shap

shap <- data.frame(
  feature = rep(c("age", "income", "tenure"), each = 40),
  shap    = c(rnorm(40, 0.3, 0.2), rnorm(40, -0.1, 0.3), rnorm(40, 0, 0.15)),
  value   = runif(120)
)
ggnext(shap, aes(shap, feature, color = value)) +
  geom_shap() + geom_vline(0, dash = "3,3") +
  labs(x = "SHAP value", y = NULL) + theme_minimal()

9.2 geom_partial_dependence

pdp <- data.frame(
  x = rep(1:10, 8), id = rep(1:8, each = 10),
  pred = as.vector(sapply(1:8, function(i) (1:10) * 0.1 * i + rnorm(10, 0, .15)))
)
ggnext(pdp, aes(x, pred, group = id)) +
  geom_partial_dependence() + theme_minimal()

9.3 geom_roc

sc <- runif(300)
roc_d <- data.frame(score = sc, truth = rbinom(300, 1, sc))
ggnext(roc_d, aes(score = score, truth = truth)) +
  geom_roc() + theme_minimal()

A perfect classifier gives an exact staircase:

perfect <- compute_stat(stat_roc(), list(
  truth = c(1, 1, 0, 0), score = c(0.9, 0.8, 0.2, 0.1),
  group = rep("all", 4)
))
data.frame(fpr = perfect$x, tpr = perfect$y)

#>   fpr tpr
#> 1 0.0 0.0
#> 2 0.0 0.5
#> 3 0.0 1.0
#> 4 0.5 1.0
#> 5 1.0 1.0
#> 6 0.0 0.0
#> 7 1.0 1.0

9.4 geom_calibration

pr <- runif(400)
cal <- data.frame(pred = pr, obs = rbinom(400, 1, pr^1.3))
ggnext(cal, aes(pred, obs)) + geom_calibration() + theme_minimal()

9.5 geom_lift_gain

s2 <- runif(250)
lg_d <- data.frame(score = s2, y = rbinom(250, 1, s2))
ggnext(lg_d, aes(score = score, truth = y)) +
  geom_lift_gain(type = "gain") + theme_minimal()

ggnext(lg_d, aes(score = score, truth = y)) +
  geom_lift_gain(type = "lift") + theme_minimal()

9.6 geom_confusion_matrix

cm <- data.frame(
  predicted = c(rep("cat", 14), rep("dog", 9), rep("bird", 6)),
  actual = c(rep("cat", 11), rep("dog", 3), rep("dog", 7),
             rep("cat", 2), rep("bird", 5), "cat")
)
ggnext(cm, aes(predicted, actual)) + geom_confusion_matrix() +
  theme(legend_position = "none")

9.7 geom_residual

m <- lm(dist ~ speed, cars)
res <- data.frame(fitted = fitted(m), resid = resid(m))
ggnext(res, aes(fitted, resid)) + geom_residual() + theme_minimal()

9.8 geom_learning_curve

lc <- data.frame(
  n = rep(c(50, 100, 200, 400, 800), 2),
  score = c(.72, .80, .85, .88, .90, .66, .75, .81, .85, .88),
  split = rep(c("train", "validation"), each = 5)
)
ggnext(lc, aes(n, score, color = split)) +
  geom_learning_curve() + theme_minimal()

9.9 geom_silhouette

sil <- data.frame(
  cluster = rep(c("1", "2", "3"), each = 25),
  width = c(runif(25, .3, .9), runif(25, .1, .7), runif(25, -.1, .6))
)
ggnext(sil, aes(width, cluster, color = cluster)) +
  geom_silhouette() + theme(legend_position = "none")

9.10 geom_embedding

emb <- data.frame(d1 = c(rnorm(40), rnorm(40, 4)),
                  d2 = c(rnorm(40), rnorm(40, 3)),
                  cluster = rep(c("a", "b"), each = 40))
ggnext(emb, aes(d1, d2, color = cluster)) +
  geom_embedding() + theme_minimal()

9.11 geom_decision_boundary

gr <- expand.grid(x = seq(0, 1, 0.04), y = seq(0, 1, 0.04))
gr$cls <- ifelse(gr$x + gr$y > 1, "a", "b")
ggnext(gr, aes(x, y, color = cls)) +
  geom_decision_boundary() + theme_minimal()