scales scales website

CRAN status R-CMD-check Codecov test coverage

One of the most difficult parts of any graphics package is scaling, converting from data values to perceptual properties. The inverse of scaling, making guides (legends and axes) that can be used to read the graph, is often even harder! The scales packages provides the internal scaling infrastructure used by ggplot2, and gives you tools to override the default breaks, labels, transformations and palettes.

Installation

# Scales is installed when you install ggplot2 or the tidyverse.
# But you can install just scales from CRAN:
install.packages("scales")

# Or the development version from Github:
# install.packages("pak")
pak::pak("r-lib/scales")

Usage

Breaks and labels

The most common use of the scales package is to control the appearance of axis and legend labels. Use a break_ function to control how breaks are generated from the limits, and a label_ function to control how breaks are turned in to labels.

library(ggplot2)
library(dplyr, warn.conflicts = FALSE)
library(lubridate, warn.conflicts = FALSE)

txhousing %>%
  mutate(date = make_date(year, month, 1)) %>%
  group_by(city) %>%
  filter(min(sales) > 5e2) %>%
  ggplot(aes(date, sales, group = city)) +
  geom_line(na.rm = TRUE) +
  scale_x_date(
    NULL,
    breaks = scales::breaks_width("2 years"),
    labels = scales::label_date("'%y")
  ) +
  scale_y_log10(
    "Total sales",
    labels = scales::label_number(scale_cut = scales::cut_short_scale())
  )

A line plot created with ggplot2, showing property sales in Texas. The x scale uses `scales::break_width()` to place breaks every second year, and `scales::label_date()` to create a custom format for the labels. The y-scale uses `scales::label_number()` to reformat the labels with `scales::cut_short_scale()`.

economics %>%
  filter(date < ymd("1970-01-01")) %>%
  ggplot(aes(date, pce)) +
  geom_line() +
  scale_x_date(NULL,
    breaks = scales::breaks_width("3 months"),
    labels = scales::label_date_short()
  ) +
  scale_y_continuous("Personal consumption expenditures",
    breaks = scales::breaks_extended(8),
    labels = scales::label_dollar()
  )

A line plot created with ggplot2, showing personal expenses between 1967 and 1970. The x axis uses `scales::break_width()` to put a break every 3 months and `scales::label_date_short()` to only show the year on the first occuring break of that year. The y axis uses `scales::breaks_extended()` to request 8 breaks, though only 6 are ultimately provided, and `scales::label_dollar()` to format the label as a dollar value.

Generally, I don’t recommend running library(scales) because when you type (e.g.) scales::label_ autocomplete will provide you with a list of labelling functions to jog your memory.

Advanced features

Scales colour palettes are used to power the scales in ggplot2, but you can use them in any plotting system. The following example shows how you might apply them to a base plot.

library(scales)
# pull a list of colours from any palette
pal_viridis()(4)
#> [1] "#440154FF" "#31688EFF" "#35B779FF" "#FDE725FF"

# use in combination with baseR `palette()` to set new defaults
palette(pal_brewer(palette = "Set2")(4))
par(mar = c(5, 5, 1, 1))
plot(Sepal.Length ~ Sepal.Width, data = iris, col = Species, pch = 20)

A scatterplot created with base plot showing the relationship between sepal length and sepal width in the Iris dataset. The points are coloured according to species and the `scales::pal_brewer()` are used to provide the colours.

scales also gives users the ability to define and apply their own custom transformation functions for repeated use.

# use new_transform to build a new transformation
transform_logp3 <- new_transform(
  name = "logp",
  transform = function(x) log(x + 3),
  inverse = function(x) exp(x) - 3,
  breaks = log_breaks()
)

set.seed(1234) # Ensures same sampling
dsamp <- sample_n(diamonds, 100)
ggplot(dsamp, aes(carat, price, colour = color)) +
  geom_point() +
  scale_y_continuous(trans = transform_logp3)

A scatterplot created with ggplot2 showing the relationship between diamond price and its carat for a subset of the data in the diamonds dataset. The y scale uses a custom log transform created with `scales::new_transform()`.