ggfoundry

Motivation

ggfoundry was inspired a little by Stack Overflow posts seeking specific shapes. But, in truth, mostly by a personal interest in getting acquainted with grid graphics (the underpinnings of ggplot2).

Shape landscape

Yes, there is already a seemingly near-infinite number of shapes out there:

• Those familiar to ggplot users (some fillable) as described in the ggplot2 documentation;
• Colourable unicodes and icons like fontawesome;
• ggimage enables the use of whole pictures;
• And then there is the DIY (Do-It-Yourself) approach: Conjuring up grobs (grid graphical objects); perhaps with a sprinkle of trigonometry.

But sometimes you just can’t find what you want. Nor manipulate it in the way you would like.

ggfoundry offers arbitrary hand-crafted colourable and fillable shapes for ggplot2 and is reviewed side-by-side with other options in contrast with alternatives.

Foundry process

These artisanal symbols begin life as vector images with two layers: an outline and a fill. Some are hand-drawn; others, such as the “bowl” set, are constructed programmatically. Each SVG pair is converted to Cairo graphics format, forged at extreme temperatures into objects of class “Picture”, and finally delicately cast as a gTree representation of the original shape. But not quite back to where we started, because they are now editable.

When cooled and finely burnished, the gTree and all its grob children may then be manipulated by geom_casting() to render the desired ggplot with those special high-end adornments.

Available shapes

ggfoundry may well be the destination of “last resort”!

After travelling the mountains, seas and forests of the world in search of that elusive shape (or small set), a hand-made grob may be the fillable “Holy Grail” sought via a Github issue.

These sets are included with the latest version of the package. You can “mix and match” shapes from different sets; the “set” is for grouping shapes in the documentation and for use in shapes_cast() to filter for the desired shapes.

library(ggfoundry)
library(dplyr)
library(forcats)
library(stringr)

df <- shapes_cast() |> 
  filter(!str_ends(shape, "3|4|5|6")) |> 
  mutate(x = row_number(), shape = fct_inorder(shape), .by = set)

df |> 
  ggplot(aes(x, set)) +
  geom_text(aes(label = shape), nudge_y = -0.5, colour = "grey40", size = 3) +
  geom_casting(aes(shape = shape), size = 0.11, fill = "skyblue") +
  scale_shape_manual(values = as.character(df$shape)) +
  scale_x_continuous(expand = expansion(add = 0.5)) +
  scale_y_discrete(expand = expansion(add = 0.7)) +
  labs(x = NULL, y = NULL, caption = "sunflowers 1-8 available") +
  theme_minimal() +
  theme(
    text = element_text(colour = "grey70"),
    axis.text.y = element_text(angle = 90, hjust = 0.5),
    axis.text.x = element_blank(),
    axis.ticks.x = element_blank(),
    legend.position = "none"
    )

Simple example

Using some made-up data simulating a “random walk”, geom_casting() adds a layer of custom shapes to the plot.

When the shape is mapped to a variable, then scale_shape_manual() is required to explicitly name the desired shapes as a character vector. This is because standard shapes (as used for example in geom_point()) are associated with a number, e.g. a circle is 19, whereas geom_casting() shapes are associated only with character strings.

Each shape’s appearance is resolved observation by observation, so aesthetics such as colour, fill, alpha, size and angle may vary continuously across the data, whether grouping is implicit or explicitly shared with group = 1.

Using scale_colour_manual() and scale_fill_manual(), we can also select a custom palette for the shape colours and fills.

# Toy Data
set.seed(123)

random_walk <- \(x, y, z) cumsum(rnorm(x, mean = y, sd = sqrt(z)))

df <- data.frame(
  x = rep(1:10, 2),
  y = c(
    random_walk(10, 1, 1),
    random_walk(10, 3, 1.3)
  ),
  group = factor(c(rep(1, 10), rep(2, 10)))
)

# Plot with geom_casting()
df |>
  ggplot(aes(x, y, shape = group, colour = group, fill = group)) +
  geom_line(show.legend = FALSE) +
  geom_casting() +
  scale_colour_manual(values = c("darkred", "darkgreen")) +
  scale_fill_manual(values = c("pink", "lightgreen")) +
  scale_shape_manual(values = c("cross1", "cross2")) +
  labs(title = "ggfoundry") +
  theme_bw() +
  theme(plot.subtitle = element_text(size = 10))

Porridge bowls

The “bowl” set brings three containers to the table: bowl0, bowl1 and bowl2, distinguished only by zero, one or two wisps of steam rising from the porridge. The bowl body and the steam share the fill, while the bowl and porridge outline take the colour. All three variants share the same canvas and bowl position, so they may be swapped freely with scale_shape_manual().

A continuous fill maps directly to any numeric variable, with no grouping workaround required. And a fill of "transparent" hides the steam while the bowl outline remains visible.

bowls <- data.frame(
  x = 1:7,
  temperature = c(5, 18, 32, 47, 63, 81, 96)
)

bowls |>
  ggplot(
    aes(
      x, 1, fill = temperature,
      shape = cut(temperature, 3, labels = c("Cold", "Just right", "Hot"))
    )
  ) +
  geom_casting(colour = "#272626", size = 0.72) +
  geom_text(aes(y = 0.35, label = temperature), size = 4) +
  scale_shape_manual(
    values = c("bowl0", "bowl1", "bowl2"),
    guide = guide_legend(
      override.aes = list(fill = c("#240691", "#C8437B", "#F8DF25")),
      keyheight = unit(2.5, "lines"), default.unit = "line"
    )
  ) +
  scale_fill_viridis_c(
    limits = c(0, 100), begin = 0.04, end = 0.94, option = "plasma",
    guide = "none"
  ) +
  scale_x_continuous(limits = c(0.35, 7.65), expand = expansion(mult = 0)) +
  scale_y_continuous(limits = c(0, 1.85), expand = expansion(mult = 0)) +
  labs(title = "Hot, just right or cold", shape = "Serving") +
  theme_void() +
  theme(legend.position = "bottom")

See the showcase article to explore other use cases and contrast with alternatives to review against other options.

Coffee and autumn

Six newcomers arrive with a new “food” set: a cup, a mug and a takeaway cup join the “container” set, while a coffeebean, a pumpkin and a jackolantern found the “food” set. As with the bowls, cup and mug steam shares the fill, so a transparent fill hides the steam while the outline remains; the takeaway sleeve follows the fill, and the ribs, bean crease, stems and carved faces follow the colour. They are different vessels, not an ordinal temperature classification.

cafe <- data.frame(
  x = 1:6,
  roast = c(8, 24, 42, 60, 78, 95),
  vessel = c("cup", "mug", "takeaway", "coffeebean", "pumpkin", "jackolantern")
)

cafe |>
  ggplot(aes(x, 1, fill = roast, shape = vessel)) +
  geom_casting(colour = "#33302D", size = 0.34) +
  scale_shape_manual(
    values = setNames(cafe$vessel, cafe$vessel),
    breaks = cafe$vessel,
    guide = guide_legend(
      override.aes = list(
        fill = c("#1B0C42", "#59106E", "#9B2964", "#D74B3F", "#F88A0C", "#F7D441")
      ),
      nrow = 2, keyheight = unit(2, "lines"), default.unit = "line"
    )
  ) +
  scale_fill_viridis_c(
    limits = c(0, 100), begin = 0.04, end = 0.94, option = "inferno",
    guide = "none"
  ) +
  scale_x_continuous(limits = c(0.35, 6.65), expand = expansion(mult = 0)) +
  scale_y_continuous(limits = c(0.55, 1.65), expand = expansion(mult = 0)) +
  labs(title = "Coffee and autumn; continuous fill, independent outline colour") +
  theme_void() +
  theme(legend.position = "bottom")

Acknowledgements

Without the pivotal grConvert (Potter 2024) and grImport2 (Potter and Murrell 2023) packages, the foundry process would not have been viable. And only thanks to the work behind ggplot2 (Wickham 2016), may the shapes cast be so beautifully visualised.