ggplot2 completely custom legend?

Question

Is there any way that I can create a completely custom legend that won't care about the aesthetics or anything else in my plot? I would like if possible to design everything from scratch. How many groups of legends I will have, what will be the title, shape, color, size, linetype, fill, label, order etc for each legend. I have already spent almost two working days try to figure out how to create the legends to look the way I want them to look (The plot itself didn't take more than a few minutes after I had the data).

Take a look at the following sample code (random data, but fine for demonstrating what I want):

require(dplyr)
require(RColorBrewer)

col <- brewer.pal(3, "Spectral")

a <- data.frame(multiplier = c(0.5, 0.7, 1.0), multiplier2 = c(0.3, 0.1), random = runif(3 * 500))
a$result = a$multiplier * a$random * runif(length(a$random)) * a$multiplier2

a_grouped_by_multiplier = group_by(a, multiplier)
means = summarise(a_grouped_by_multiplier, mean_rand = mean(random), mean_res = mean(result))

ggplot(a, aes(x = random, y = result)) +
  geom_density2d(bins = 20) +
  geom_point(aes(color = factor(multiplier2)), size = 0.7) +
  geom_vline(data = means, aes(xintercept = mean_rand), color = "orange", linetype = "solid", size = 1.5, show.legend = TRUE) +
  geom_hline(data = means, aes(yintercept = mean_res), color = "red", linetype = "dashed", size = 1.5, show.legend = TRUE) +
  scale_color_manual(name = "Values", values = c(col[1], col[2]), labels = c("* 0.1", "* 0.3")) +
  facet_grid(~ multiplier) +
  theme(panel.grid.major = element_line(colour = "white", linetype = "dashed", size = 0.3),
        panel.grid.minor = element_blank(),
        panel.background = element_rect(fill = "#555555"),
        legend.key = element_rect(fill = "#555555"))

That creates the following plot:

I have tried to customize endless parameters to get the desired result. Used all the different scale_*_manual functions with different parameters, used the show.legends = FALSE, tried the guide() function with different guide_legend() parameters, I tried to make the color, linetype, size etc parameters part of the aesthetics (one by one and all combined), but nothing so far worked for creating a legend like the following (created by inkscape here):

My first problem is that I cannot get two legend groups: one for the "Values" and one for the "Averages". Second problem is that due to the geom_hline and geom_vline, vertical and horizontal lines appear in all of the legend boxes. Note that I also use a different data frame for the geom_hline and geom_vline.

Solution

I managed to plot what I wanted but in an ugly way when it comes to coding. In short, I added different aesthetics until I got at least two legend groups with the desired number of elements and shapes. Then I used the scale_*_manual functions and the guide() function to override some of the aesthetics and make the legend look almost as it should (still I didn't manage to have a vertical line in a legend box).

The code is the following:

ggplot(a, aes(x = random, y = result)) +
  geom_density2d(bins = 20) +
  geom_point(aes(color = factor(multiplier2), shape = factor(multiplier2), fill = factor(multiplier2)), stroke = 0.01, size = 2.5) +
  geom_vline(data = means, aes(xintercept = mean_rand, color = "mean_rand", size = "mean_rand"), show.legend = FALSE) +
  geom_hline(data = means, aes(yintercept = mean_res, color = "mean_res", size = "mean_res"), linetype = 12, show.legend = FALSE) +
  scale_color_manual(values = c(col[1], col[2], "orange", "red")) +                             # Provides the desired colors for the plot
  scale_shape_manual(name = "Values", values = c(21, 22), labels = c("*0.1", "*0.3")) +         # Provides the desired shape and plots the "Values" legend
  scale_fill_manual(name = "Values", values = c(col[1], col[2]), labels = c("*0.1", "*0.3")) +  # Provides the fill for the shapes and merges the legend with the shape ("Values") legend
  scale_size_manual(name = "Averages", values = c(1.5, 1.5), labels = c("Random", "Result")) +  # Provides a legend for the Averages but looks ugly.....
  guides(color = FALSE,  # Hides the "color" legend because it has 4 values and we don't want that.
         shape = guide_legend(order = 1), # Forces the "Values" legend group to be on top
         fill = guide_legend(order = 1),  # Forces the "Values" legend group to be on top
         size = guide_legend(order = 2, override.aes = list(linetype = c(1, 12), color = c("orange", "red")))) +  # Makes the "Average" legend to look as it should, by overriding the aesthetics
  facet_grid(~ multiplier) +
  theme_dark() +
  theme(panel.grid.major = element_line(colour = "white", linetype = "dashed", size = 0.2),
        panel.background = element_rect(fill = "#555555"),
        legend.key = element_rect(fill = "#555555"))

And that's how the result looks like:

Comments

[alistaire]

The short answer is no, ggplot legends are inexorably tied to aesthetics. They are very customizable, though. In this case, you need to find a way to make the averages an aesthetic, maybe by making a dummy data.frame and using geom_abline.