ggblanket

David Hodge

Overview

ggblanket is a package of ggplot2 wrapper functions.

The primary objective is to simplify ggplot2 visualisation.

Secondary objectives relate to:

• Design: produce well-designed visualisation by default
• Scope: cover the most useful 80% of what ggplot2 does
• Alignment: use conventions generally aligned with ggplot2.

How it works

1. Over thirty gg_* wrapper functions
2. A single col argument to colour and fill by a variable
3. A pal argument to customise colours
4. A facet argument to facet by a variable
5. An additional facet2 argument to facet by a 2nd variable
6. Conversion of unspecified titles to sentence case
7. Ability to customise by prefixed arguments
8. Ability to customise the theme
9. Access to other geom_* arguments via …
10. Pretty continuous positional scales
11. Ability to add multiple geom_* layers
12. Other key differences to ggplot2

library(dplyr)
library(ggplot2)
library(ggblanket)
library(patchwork)

penguins2 <- palmerpenguins::penguins |>
  mutate(sex = stringr::str_to_sentence(sex)) |>
  tidyr::drop_na(sex)

1. Over thirty gg_* wrapper functions

Each gg_* function wraps a ggplot2 ggplot(aes(...)) function with the applicable ggplot2 geom_*() function. Each gg_* function is named after the geom_* function they wrap.

# ggplot2
penguins2 |> 
  ggplot() + 
  geom_point(aes(x = flipper_length_mm, 
                 y = body_mass_g))

# ggblanket
penguins2 |>
  gg_point(
    x = flipper_length_mm,
    y = body_mass_g)

2. A single col argument to colour and fill by a variable

The colour and fill aesthetics of ggplot2 are merged into a single concept represented by the col argument. This argument means that everything should be coloured according to it, i.e. all points, lines and polygon interiors.

# ggplot2
p1 <- penguins2 |> 
  ggplot() + 
  geom_point(aes(x = flipper_length_mm, 
                 y = body_mass_g,
                 colour = species)) 

p2 <- penguins2 |>
  ggplot() +
  geom_density(aes(x = flipper_length_mm, 
                   fill = species)) +
  labs(fill = "Species")

p1 / p2

# ggblanket
p1 <- penguins2 |>
  gg_point(
    x = flipper_length_mm,
    y = body_mass_g, 
    col = species)

p2 <- penguins2 |>
  gg_density(
    x = flipper_length_mm, 
    col = species)

p1 / p2

3. A pal argument to customise colours

The pal argument is used to customise the colours of the geom. A user can provide a vector of colours to this argument. It can be named or not. It works in a consistent way - regardless of whether a col argument is added or not. A named palette can be used to make individual colours stick to particular values. ggblanket uses alpha defaults to make outputs look pretty.

# ggplot2
p1 <- penguins2 |>
  ggplot() +
  geom_histogram(aes(x = body_mass_g),
                 fill = "#414D6B")

p2 <- penguins2 |>
  ggplot() +
  geom_jitter(aes(x = species, 
                  y = body_mass_g, 
                  colour = sex)) +
  scale_colour_manual(values = c("#1B9E77", "#9E361B"))

p1 / p2

# ggblanket
p1 <- penguins2 |>
  gg_histogram(
    x = body_mass_g, 
    pal = "#414D6B")

p2 <- penguins2 |>
  gg_jitter(
    x = species, 
    y = body_mass_g, 
    col = sex, 
    pal = c("#1B9E77", "#9E361B"))

p1 / p2

4. A facet argument to facet by a variable

Faceting is treated as if it were an aesthetic. Users just provide an unquoted variable to facet by. If a single facet (or facet2) variable is provided, it’ll default to a “wrap” layout. But users can change this with a facet_layout = "grid" argument.

# ggplot2
penguins2 |>
  ggplot() +
  geom_violin(aes(x = sex, 
                  y = body_mass_g)) +
  facet_wrap(vars(species)) 

# ggblanket
penguins2 |>
  gg_violin(
    x = sex,
    y = body_mass_g,
    facet = species)

5. An additional facet2 argument to facet by a 2nd variable

A facet2 argument is also provided for extra functionality and flexibility. If both facetand facet2 variables are provided, then it’ll default to a “grid” layout of facet by facet2. But users can change this with a facet_layout = "wrap" argument.

# ggplot2
penguins2 |>
  ggplot() +
  geom_histogram(aes(x = flipper_length_mm)) +
  facet_grid(rows = vars(sex), cols = vars(species))

# ggblanket
penguins2 |>
  gg_histogram(
    x = flipper_length_mm,
    facet = species,
    facet2 = sex)

6. Conversion of unspecified titles to sentence case

Unspecified x, y, and col titles are converted to sentence case with snakecase::to_sentence. All titles can be manually changed using the *_title arguments. The default conversion is intended to make titles sometimes able to be left as is. Use *_title = "" to remove a title.

# ggplot2
penguins2 |>
  ggplot() +
  geom_point(aes(x = flipper_length_mm, 
                 y = body_mass_g, 
                 colour = sex)) +
  facet_wrap(vars(species)) +
  scale_x_continuous(breaks = scales::breaks_pretty(n = 3)) 

# ggblanket
penguins2 |>
  gg_point(
      x = flipper_length_mm,
      y = body_mass_g, 
      col = sex,
      facet = species)

7. Ability to customise by prefixed arguments

Prefixed arguments are available to customise titles, scales, guides, and faceting. These prefixes organise the adjustments by whether they relate to x, y, col or facet.

# ggplot2
penguins2 |>
  ggplot() +
  geom_jitter(aes(x = species, 
                  y = body_mass_g, 
                  colour = sex)) +
  expand_limits(y = 0) +
  scale_x_discrete(labels = \(x) stringr::str_sub(x, 1, 1)) +
  scale_y_continuous(breaks = scales::breaks_width(1500),
                     labels = scales::label_number(big.mark = " "),
                     expand = expansion(mult = c(0, 0.05)),
                     trans = "sqrt") +
  labs(x = "Species", y = "Body mass (g)", col = NULL) +
  theme(legend.position = "top") +
  theme(legend.justification = "left") +
  scale_colour_manual(values = scales::hue_pal()(2), 
                      guide = ggplot2::guide_legend(title.position = "top"))

# ggblanket
penguins2 |>
  gg_jitter(
    x = species,
    y = body_mass_g,
    col = sex,
    x_labels = \(x) stringr::str_sub(x, 1, 1),
    y_include = 0,
    y_breaks = scales::breaks_width(1500), 
    y_labels = scales::label_number(big.mark = " "), 
    y_expand = expansion(mult = c(0, 0.05)),
    y_trans = "sqrt",
    y_title = "Body mass (g)", 
    col_legend_place = "top", 
    col_title = "")

These prefixed arguments work nicely with the Rstudio autocomplete, if users:

• ensure their settings support the use of tab for autocompletions and multi-line autocompletions (i.e. Tools - Global Options - Code - Completion)
• pipe data into the gg_* functions.

With these settings and use of the pipe, users can type the prefix, and then use the tab and arrow keys to assist in finding and selecting the arguments they need to adjust.

8. Ability to customise the theme

Users can use the theme argument in a gg_* function for the theme of a plot.

Alternatively, users can set the theme globally using the ggplot2::theme_set function, such that all subsequent plots will use this by default.

ggblanket provides two complete ggplot2 theme functions called light_mode (the default) and dark_mode. The first argument is the base_size. This changes the size of all the text to this, except the title is 10% higher and the caption is 10% lower. In quarto, it is likely that users will want to set the *_mode theme to have a larger base_size (e.g. ggplot2::theme_set(light_mode(11))).

Note that theme_set(theme_grey()) resets the set theme for ggplot2 code to theme_grey and for ggblanket gg_* functions to light_mode(). If you want ggblanket gg_* functions to default to using theme_grey(), then you must modify the base_size slightly (e.g. theme_set(theme_grey(11.01))).

# ggblanket

# theme_set(dark_mode(10))

penguins2 |>
  gg_point(
    x = flipper_length_mm,
    y = body_mass_g,
    col = sex,
    title = "Penguins body mass by flipper length",
    subtitle = "Palmer Archipelago, Antarctica",
    caption = "Source: Gorman, 2020", 
    theme = dark_mode(10))

Note that the gg_* function will by default adjust what gridlines are present and the placement of the legend. Therefore, if you are providing a theme other than light_mode or dark_mode, ggblanket works well if this theme has both vertical and horizontal gridlines. If users want everything adjusted as per the theme, then they can + their theme onto the plot instead.

# ggblanket
p1 <- penguins2 |>
  gg_point(
    x = flipper_length_mm,
    y = body_mass_g,
    col = sex,
    x_breaks = scales::breaks_pretty(n = 3),
    theme = theme_grey(),
    title = "theme= theme_grey()")

p2 <- penguins2 |>
  gg_point(
    x = flipper_length_mm,
    y = body_mass_g,
    col = sex,
    x_breaks = scales::breaks_pretty(n = 3),
    title = "+ theme_grey()") +
  theme_grey()

p1 + p2

9. Access to other geom_* arguments via ...

The ... argument is placed in the gg_* function within the wrapped ggplot2::geom_* function. This means all other arguments in the geom_* function are available to users. Common arguments from ... to add are size, linewidth and width.

# ggblanket
penguins2 |>
  gg_smooth(
    x = flipper_length_mm,
    y = body_mass_g,
    col = sex, 
    linewidth = 0.5, #accessed via geom_smooth
    level = 0.99) #accessed via geom_smooth

10. Pretty continuous positional scales

Where the orientation is normal (i.e. vertical):

• Default x scale limits and expanding are as per ggplot2 defaults
• Default continuous y scale limits (that are non-transformed) start and end on a break with zero expanding
• Default categorical (or numeric transformed) y scale limits and expanding are as per ggplot2 defaults
• Default removal of vertical gridlines

It does the opposite where the orientation is horizontal.

Note this symmetry approach does not apply: * if a transformation other than identity or reverse is applied to x or y scales. * for gg_raster, gg_contour_filled or gg_density_2d_filled

In some circumstances the ggplot2 approach to default scales may be preferable. In these cases, users can revert to the ggplot2 approach by using *_limits = c(NA, NA) and *_expand = c(0.05, 0.05) (or add scale_*_continuous()).

# ggplot2
penguins2 |>
  group_by(species, sex) |> 
  summarise(body_mass_g = mean(body_mass_g)) |> 
  ggplot() +
  geom_col(aes(x = body_mass_g, 
               y = species, 
               fill = sex),
           position = "dodge",
           width = 0.66)

# ggblanket
penguins2 |>
  group_by(species, sex) |> 
  summarise(body_mass_g = mean(body_mass_g)) |> 
  gg_col(
    x = body_mass_g,
    y = species,
    col = sex,
    position = "dodge",
    width = 0.66)

Sometimes with small plots or faceted plots etc, the labels can be too squashed. Making the breaks width bigger can waste space, due to the afore-mentioned approach of ggblanket to making pretty scales. An alternative approach is to use the str_keep_seq function with the *_labels arguments to only keep every 2nd (or nth) label.

# ggblanket
penguins2 |>
  group_by(species, sex) |> 
  summarise(body_mass_g = mean(body_mass_g)) |> 
  ungroup() |> 
  gg_col(
    y = body_mass_g,
    x = species,
    col = sex,
    position = "dodge",
    width = 0.5,
    x_labels = \(x) stringr::str_sub(x, 1, 1),
    y_labels = \(x) str_keep_seq(x),
    title = "Keep every 2nd label", 
    theme = light_mode(title_face = "plain"))

11. Ability to add multiple geom_* layers

Users can make plots with multiple layers with ggblanket by adding on ggplot2::geom_* layers.

The gg_* function puts the aesthetic variables (i.e. x, y, col) within the wrapped ggplot function. Therefore, these aesthetics will inherit to any subsequent layers added.

Where there are multiple geom layers in a desired plot, users should determine the gg_* function with care:

1. The gg_* function should be appropriate to be the bottom layer of the plot. This is because the geoms will plot in order.

# ggblanket + ggplot2
p1 <- ggplot2::economics |>
  slice_min(order_by = date, n = 10) |> 
  gg_line(
    x = date,
    y = unemploy,
    pal = guardian()[1],
    x_title = "",
    y_title = "Unemployment",
    y_include = 0,
    linewidth = 1, 
    x_breaks = scales::breaks_width("3 months"),
    title = "gg_line + geom_point", 
    theme = light_mode(title_face = "plain")) +
  geom_point(colour = guardian()[2])

p2 <- ggplot2::economics |>
  slice_min(order_by = date, n = 10) |> 
  gg_point(
    x = date,
    y = unemploy,
    pal = guardian()[2],
    x_title = "",
    y_title = "Unemployment",
    y_include = 0,
    x_breaks = scales::breaks_width("3 months"),
    title = "gg_point + geom_line", 
    theme = light_mode(title_face = "plain")) +
  geom_line(colour = guardian()[1], linewidth = 1)

p1 + p2

2. If some geom layers have a col aesthetic and some do not, then a gg_* function should be chosen that has a col argument in it. This will enable ggblanket legend placement and access to col_* arguments. It is also a more reliable approach. If later layers do not require the col aesthetic, then the inheit.aes = FALSE argument should be used.

3. In some situations, gg_blank may be required.

Users need to ensure that the scales built by their gg_* function are appropriate for subsequent layers. Plot scales are built by the gg_* function based on the data, x, y, *_limits, *_include, stat, position and coord arguments in the gg_* function.

# ggblanket + ggplot2
d <- penguins2 |>
  group_by(species) |>
  summarise(body_mass_g = mean(body_mass_g)) |>
  mutate(lower = body_mass_g * 0.95) |> 
  mutate(upper = body_mass_g * 1.2)

p1 <- d |>
  gg_col(
    x = body_mass_g,
    y = species,
    col = species,
    width = 0.75,
    x_include = c(0, max(d$upper)), 
    x_labels = \(x) x / 1000, 
    x_title = "Body mass kg",
    col_legend_place = "none") +
  geom_errorbar(aes(xmin = lower, xmax = upper), 
                colour = "black",
                width = 0.1) 

p2 <- d |>
  gg_col(
    x = body_mass_g,
    y = species,
    col = species,
    colour = "#d3d3d3",
    fill = "#d3d3d3",
    width = 0.75,
    x_include = c(0, max(d$upper)), 
    x_labels = \(x) x / 1000, 
    x_title = "Body mass kg",
    col_legend_place = "none") +
  geom_errorbar(aes(xmin = lower, xmax = upper), 
                width = 0.1) 

p1 / p2

12. Other key differences to ggplot2

Unquoted variables only

ggblanket requires unquoted variables only for x, y, col, facet and facet2. You cannot wrap these in a function. Instead you need to apply the function to the relevant variable in the data prior to plotting. For example, reordering or reversing a factor or dropping NAs.

p1 <- diamonds |>
  count(color) |>
  gg_col(
    x = n,
    y = color,
    width = 0.75,
    x_labels = \(x) x / 1000,
    x_title = "Count (thousands)",
    title = "Default y",
    theme = light_mode(title_face = "plain")
  )


p2 <- diamonds |>
  count(color) |>
  mutate(color = forcats::fct_rev(color)) |>
  gg_col(
    x = n,
    y = color,
    width = 0.75,
    x_labels = \(x) x / 1000,
    x_title = "Count (thousands)",
    title = "Reverse y",
    theme = light_mode(title_face = "plain")
  )

p3 <- diamonds |>
  count(color) |>
  mutate(color = forcats::fct_reorder(color, n)) |>
  gg_col(
    x = n,
    y = color,
    width = 0.75,
    x_labels = \(x) x / 1000,
    x_title = "Count (thousands)",
    title = "Reordered y ascending by x",
    theme = light_mode(title_face = "plain")
  )

p4 <- diamonds |>
  count(color) |>
  mutate(color = color |>
           forcats::fct_reorder(n) |>
           forcats::fct_rev()) |>
  gg_col(
    x = n,
    y = color,
    width = 0.75,
    x_labels = \(x) x / 1000,
    x_title = "Count (thousands)",
    title = "Reordered y decending by x",
    theme = light_mode(title_face = "plain")
  )

(p1 + p2) / (p3 + p4)

Unused factor levels kept

ggblanket keeps unused factor levels in the plot. If users wish to drop unused levels they should likewise do it in the data prior to plotting.

p1 <- diamonds |> 
  count(color) |> 
  filter(color %in% c("E", "G", "I")) |>
  gg_point(
    x = n,
    y = color,
    x_labels = \(x) x / 1000,
    x_title = "Count (thousands)",
    title = "A factor filtered",
    theme = light_mode(title_face = "plain"))

p2 <- diamonds |> 
  count(color) |>
  filter(color %in% c("E", "G", "I")) |>
  mutate(color = forcats::fct_drop(color)) |> 
  gg_point(
    x = n,
    y = color,
    x_labels = \(x) x / 1000,
    x_title = "Count (thousands)",
    title = "A factor filtered & unused levels dropped",
    theme = light_mode(title_face = "plain"))

p1 + p2

Different pal and alpha defaults

ggblanket uses different defaults for colouring. The default pal is:

• #357BA2 or mako(9)[5] for no col variable
• ggblanket::guardian for a discrete col variable with 4 or less levels (or unique values if a character). This palette is colourblind safe.
• scales::hue_pal for a discrete col variable with 5 or more levels (or unique values if not ordered).
• viridis::mako reversed for a continuous col variable
• "#bebebe" or grey for NA

ggblanket uses different alpha defaults for the different gg_* functions. Polygons that generally have no gap or overlap default to 1: gg_bin_2d, gg_contour_filled, gg_density_2d_filled, gg_hex - as well as gg_sf for polygons with a col aesthetic. Polygons that generally overlap default to 0.5: gg_density. Polygons that generally have key lines within them also default to 0.5: gg_boxplot, gg_crossbar, gg_ribbon and gg_smooth. gg_label defaults to 0.05. gg_blank has no alpha argument. Other polygons default to 0.9: gg_area, gg_bar, gg_col, gg_histogram, gg_polygon, gg_rect, gg_tile and gg_violin. For all other contexts, alpha defaults to 1.

Different oob & clip defaults

By default, ggblanket keeps values outside of the limits (*_oob = scales::oob_keep) in calculating the geoms and scales to plot. It also does not clip anything outside the cartesian coordinate space by default (coord = ggplot2::coord_cartesian(clip = "off")).

ggplot2 by default drops values outside of the limits in calculating the geoms and scales to plot (scales::oob_censor), and clips anything outside the cartesian coordinate space (coord = ggplot2::coord_cartesian(clip = "on")).

Users should be particularly careful when setting limits for stats other than identity.

p1 <- economics |> 
  gg_smooth(
    x = date, 
    y = unemploy, 
    y_labels = \(x) str_keep_seq(x),
    title = "No x_limits set", 
    theme = light_mode(title_face = "plain")) +
  geom_vline(xintercept = c(lubridate::ymd("1985-01-01", "1995-01-01")),
             col = guardian(n = 1), 
             linetype = 3) +
  geom_point(col = guardian(n = 1), alpha = 0.05)

p2 <- economics |> 
  gg_smooth(
    x = date, 
    y = unemploy, 
    x_limits = c(lubridate::ymd("1985-01-01", "1995-01-01")),
    x_labels = \(x) stringr::str_sub(x, 3, 4),
    y_labels = \(x) str_keep_seq(x),
    title = "x_limits set", 
    theme = light_mode(title_face = "plain")) +
  geom_point(col = guardian(n = 1), alpha = 0.1)

p3 <- economics |> 
  gg_smooth(
    x = date, 
    y = unemploy, 
    x_limits = c(lubridate::ymd("1985-01-01", "1995-01-01")),
    x_labels = \(x) stringr::str_sub(x, 3, 4),
    y_labels = \(x) str_keep_seq(x),
    coord = coord_cartesian(clip = "on"), 
    title = "x_limits set & cartesian space clipped", 
    theme = light_mode(title_face = "plain")) +
  geom_point(col = guardian(n = 1), alpha = 0.1)

p4 <- economics |> 
  gg_smooth(
    x = date, 
    y = unemploy, 
    x_limits = c(lubridate::ymd("1985-01-01", "1995-01-01")),
    x_labels = \(x) stringr::str_sub(x, 3, 4),
    x_oob = scales::oob_censor,
    y_labels = \(x) str_keep_seq(x),
    title = "x_limits set & x_oob censored", 
    theme = light_mode(title_face = "plain")) +
  geom_point(col = guardian(n = 1), alpha = 0.1)

p5 <- economics |> 
  filter(between(date, lubridate::ymd("1985-01-01"), lubridate::ymd("1995-01-01"))) |> 
  gg_smooth(
    x = date, 
    y = unemploy,
    x_labels = \(x) stringr::str_sub(x, 3, 4),
    y_labels = \(x) str_keep_seq(x),
    title = "x data filtered", 
    theme = light_mode(title_face = "plain")) +
  geom_point(col = guardian(n = 1), alpha = 0.1)

p1 / (p2 + p3) / (p4 + p5) 

Computational speed is sacrificed

ggblanket is much slower than ggplot2 in computational speed, due to the hack that underlies ggblanket to make its pretty continuous scales.

bench::mark({
  penguins2 |> 
    gg_point(x = flipper_length_mm, 
             y = body_mass_g, 
             col = species)
}, iterations = 10)
#> # A tibble: 1 × 6
#>   expression                             min median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                           <bch> <bch:>     <dbl> <bch:byt>    <dbl>
#> 1 { gg_point(penguins2, x = flipper_l… 107ms  109ms      9.22     806KB     9.22

bench::mark({
  penguins2 |>
    ggplot() +
    geom_point(aes(x = flipper_length_mm, y = body_mass_g, colour = species)) 
}, iterations = 10)
#> # A tibble: 1 × 6
#>   expression                             min median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                          <bch:> <bch:>     <dbl> <bch:byt>    <dbl>
#> 1 { ggplot(penguins2) + geom_point(a… 1.84ms 2.03ms      454.    16.3KB        0

Other minor stuff

• *_title is equivalent to ggplot2::labs(* = ...) or ggplot2::scale_*(name = ...).
• Logical variables are reversed, so that TRUE always comes before FALSE.
• *_include works in a similar way to ggplot2::expand_limits(* = ...).

Further information

See the ggblanket website for further information, including articles and function reference.