README

ggtintshade is an extension to the ggplot2 plotting library that allows for the tint/shade of a color to be mapped to an aesthetic in addition to its hue. This permits visual grouping of similar points in color space while still allowing a color legend to disambiguate them overall. It supports both nested and crossed designs.

Installation

You can install the most recent stable version of ggtintshade from CRAN (soon!) as follows:

install.packages("ggtintshade")

# install.packages("devtools")
devtools::install_github("wkumler/ggtintshade")

Examples

Crossed vs nested data

The original inspiration behind ggtintshade came from metabolomics experiments where I wanted to be able to discuss both individual compounds as well as the groups they fell into. A useful visual guide for this is to have all of one compound type be a single color, while individual compounds within that group have different shades.

library(ggtintshade)

metab_data <- data.frame(
  metab = rep(c("Alanine", "Threonine", "Glycine",
                "Glycine betaine", "Proline betaine", "Carnitine",
                "DMSP", "DMS-Ac", "Isethionate"), 3),
  metab_group = rep(rep(c("Amino acid", "Betaine", "Sulfur"), each = 3), 3),
  tripl = rep(c("A", "B", "C"), each = 9),
  area  = runif(27)
)
metab_data$metab <- factor(metab_data$metab, levels = unique(metab_data$metab))

ggplot(metab_data) +
  geom_col_tintshade(aes(x=tripl, y=area, fill=metab_group, tintshade = metab))

This is a nice example of nested data, where each individual entry belongs to a single group. ggtintshade also handles crossed data, where each shade should map into multiple groups. A good example of this is your D&D-style “alignment” chart.

align_data <- data.frame(
  alignment=1:9,
  moral=rep(c("good", "neutral", "evil"), each=3),
  meta=rep(c("lawful", "neutral", "chaotic"), length.out=9)
)
align_data$moral <- factor(align_data$moral, levels=rev(unique(align_data$moral)))
align_data$meta <- factor(align_data$meta, levels=unique(align_data$meta))

ggplot(align_data) +
  geom_raster_tintshade(aes(x=meta, y=moral, fill=meta, tintshade=moral)) +
  scale_fill_manual(breaks = c("lawful", "neutral", "chaotic"), values=c("#cca40a", "#b52060", "#7623b2")) +
  coord_equal()

That last plot also demonstrates how well ggtintshade handles normal ggplot2 behavior. Manually specified colors are handled naturally and the interaction is seamless, but you can also control the degree of lightening/darkening using the expected ggplot2 syntax for the new aesthetic with the associated scale. For example, using the diamonds dataset:

grp <- c(I1 = "I", SI2 = "SI", SI1 = "SI", VS2 = "VS", VS1 = "VS", VVS2 = "VVS", VVS1 = "VVS", IF = "IF")
diamonds$clarity_group <- factor(grp[as.character(diamonds$clarity)], levels = c("I", "SI", "VS", "VVS", "IF"))
mp <- aggregate(price ~ clarity + clarity_group, diamonds, mean)

crossed_gp <- ggplot(mp) +
  geom_col_tintshade(aes(clarity, price, fill = clarity_group, tintshade = clarity)) +
  scale_tintshade_discrete(range = c(0.4, 0.6)) +
  ggtitle("Nested diamonds") +
  theme(axis.text.x = element_text(angle=90, hjust=1, vjust=0.5))
nested_gp <- ggplot(diamonds) +
  geom_bar_tintshade(aes(x=cut, fill = cut, tintshade = clarity), color="black") +
  ggtitle("Crossed diamonds") +
  scale_tintshade_discrete(range = c(0.1, 0.9)) +
  theme(axis.text.x = element_text(angle=90, hjust=1, vjust=0.5))
crossed_gp + nested_gp

Using tintshade instead of alpha

This package is a much better way to map a lightness aesthetic than the advice commonly offered online to use alpha instead. (e.g here, here, here and here) For one, alpha only lightens, not darkens, and creates an unhelpful transparency that must be handled. It also avoids having to calculate an interaction and then set a manual scale for two aesthetics pasted together. Additionally, using an overlapping alpha trace is inefficient and can cause issues if exported to a vectorized format because each point has multiple values.

temp.data = data.frame (
  Species  = rep(c("A","B"),each=2, times=2),
  Status = rep(c("An","Bac"), times=4),
  Sex = rep(c("Male","Female"), each=4, times=1),
  Proportion = c(6.86, 7.65, 30.13, 35.71, 7.13, 10.33, 29.24, 31.09)
)
init_alpha <- ggplot(temp.data, aes(x = Species, y = Proportion, fill = Status, alpha = Species)) +
  geom_bar(stat='identity', position = position_dodge(width = 0.73), width=.67) +
  facet_grid(Sex ~ .) +
  scale_fill_manual(name = "Status", labels = c("An","Bac"), values = c("#86a681","#0a3e03")) +
  ggtitle("Without ggtintshade")
new_tinted <- ggplot(temp.data, aes(x = Species, y = Proportion, fill = Species, tintshade = Status)) +
  geom_bar_tintshade(stat='identity', position = position_dodge(width = 0.73), width=.67) +
  facet_grid(Sex ~ .) +
  scale_fill_manual(name = "Status", labels = c("An","Bac"), values = c("#cf944c","#0a3e03")) +
  scale_tintshade_discrete(range = c(0.3, 0.7)) +
  ggtitle("With ggtintshade")
init_alpha + new_tinted
#> Warning: Using alpha for a discrete variable is not advised.

d <- data.frame(
  x=rep(1:20, 5), y=rnorm(100, 5, .2) + rep(1:5, each=20), 
  z=rep(1:20, 5), grp=factor(rep(1:5, each=20))
)

init_alpha <- ggplot(d) +
  geom_path(aes(x, y, color=grp), linewidth=2, lineend=0) +
  geom_path(aes(x, y, group=grp, alpha=z), linewidth=2, lineend=0) +
  ggtitle("Without ggtintshade")
new_tinted <- ggplot(d) + 
  geom_path_tintshade(aes(x, y, color=grp, tintshade=z), linewidth=2, lineend=0) +
  scale_tintshade_continuous(range = c(0.5, 0)) +
  ggtitle("With ggtintshade")
init_alpha + new_tinted

Other examples:

mpgsub <- head(mpg, 60)
mpgsub$model <- factor(mpgsub$model, levels=unique(mpgsub$model))
ggplot(mpgsub, aes(displ, hwy, colour = manufacturer, tintshade = model)) +
  geom_point_tintshade(size = 3)

ggplot(penguins) + 
  geom_point_tintshade(aes(x=bill_len, y=bill_dep, fill=species, tintshade=sex), 
                       pch=21, color="black", size=3)
#> Warning: Removed 2 rows containing missing values or values outside the scale range
#> (`geom_point_tintshade()`).

This last example also shows how NA tintshade values are mapped to the untinted shade, which could cause some confusion. The recommended approach in this case is to ensure that the NA values are also mapped to an additional aesthetic, e.g. shape.

Internals

Internally, this is a bit of a hack. Base ggplot2 treats each aesthetic separately and does not like allowing them to interact in this way. We get around this issue first by creating a cache that’s passed around in the ggproto object and then we (ab)use the use_defaults step where all the necessary information is available. For more details about this, see the [internals vignette].

ggtintshade uses colorspace::lighten to actually modify the colors, so review the documentation there for more information about how the color values are remapped.

The individual geoms are generated via a factory [R/geoms.R] instead of copy-pasting code, so adding new geoms isn’t difficult as long as they inherit naturally from an existing ggplot2 function.