Recipe #1, geom_medians() and geom_means()

The Goal

Why write new geom_* functions? When visualizations requires computation before plotting, custom geom_() or stat_() functions can streamline your workflow. By defining new Stats objects and using them to define new geom_*() functions, you can integrate calculations directly into the plotting pipeline. In the following code, we’ll demonstrate the process to define geom_medians() to add a point at the means of x and y which can be used as follows:

ggplot(data = penguins) + 
  aes(x = bill_depth_mm,
      y = bill_length_mm) + 
  geom_point() + 
  geom_means(size = 8, color = "red") # new function!

In this exercise, we’ll demonstrate how to define the new extension function geom_medians() to add a point at the medians x and y. Then you’ll be prompted to define geom_means() based on what you’ve learned.

Step 00: Loading packages and prepping data

Handling missingness is not a discussion of this tutorial, so we’ll only use complete cases.

library(tidyverse)
library(palmerpenguins)
penguins_clean <- remove_missing(penguins) 
glimpse(penguins_clean)
Rows: 333
Columns: 8
$ species           <fct> Adelie, Adelie, Adelie, Adelie, Adelie, Adelie, Adel…
$ island            <fct> Torgersen, Torgersen, Torgersen, Torgersen, Torgerse…
$ bill_length_mm    <dbl> 39.1, 39.5, 40.3, 36.7, 39.3, 38.9, 39.2, 41.1, 38.6…
$ bill_depth_mm     <dbl> 18.7, 17.4, 18.0, 19.3, 20.6, 17.8, 19.6, 17.6, 21.2…
$ flipper_length_mm <int> 181, 186, 195, 193, 190, 181, 195, 182, 191, 198, 18…
$ body_mass_g       <int> 3750, 3800, 3250, 3450, 3650, 3625, 4675, 3200, 3800…
$ sex               <fct> male, female, female, female, male, female, male, fe…
$ year              <int> 2007, 2007, 2007, 2007, 2007, 2007, 2007, 2007, 2007…

Step 0: use base ggplot2 to get the job done

It’s a good idea to get things done without Stat extension first, just using ‘base’ ggplot2. The computational moves you make here can serve a reference for building our extension function.

# Compute.
penguins_medians <- penguins_clean |> 
  summarize(bill_length_mm_median = median(bill_length_mm),
            bill_depth_mm_median = median(bill_depth_mm))

# Plot.
penguins_clean |> 
  ggplot() + 
  aes(x = bill_depth_mm, y = bill_length_mm) + 
  geom_point() + 
  geom_point(data = penguins_medians,
             aes(x = bill_depth_mm_median,
                 y = bill_length_mm_median),
             size = 8, color = "red") + 
  labs(title = "Created with base ggplot2")

Use ggplot2::layer_data() to inspect the render-ready data internal in the plot. Your Stat will help prep data to look something like this.

layer_data(plot = last_plot(), 
           i = 2) # layer 2, the computed means, is of interest
     x    y PANEL group shape colour size fill alpha stroke
1 17.3 44.5     1    -1    19    red    8   NA    NA    0.5

Step 1: Define compute. Test.

Now you are ready to begin building your extension function. The first step is to define the compute that should be done under-the-hood when your function is used. We’ll define this in a function called compute_group_medians(). The data input will look similar to the plot data. You will also need to include a scales argument, which ggplot2 uses internally.

Define compute.

# Define compute.
compute_group_medians <- function(data, scales){ 
  data |> 
    summarize(x = median(x),
              y = median(y))
}
NoteYou may have noticed …

  1. … the scales argument in the compute definition, which is used internally in ggplot2. While it won’t be used in your test (up next), you do need so that the computation will work in the ggplot2 setting.

  2. … that the compute function can only be used with data with variables x and y. These aesthetic variables names, relevant for building the plot, are generally not found in the raw data inputs for plot.

Test compute.

# Test compute. 
penguins_clean |>
  select(x = bill_depth_mm,  
         y = bill_length_mm) |>  
  compute_group_medians()
# A tibble: 1 × 2
      x     y
  <dbl> <dbl>
1  17.3  44.5
NoteYou may have noticed …

… that we prepare the data to have columns with names x and y before testing. Computation will fail if variables x and y are not present given the function’s definition. In a plotting setting, columns are renamed by mapping aesthetics, e.g. aes(x = bill_depth, y = bill_length).

Step 2: Define new Stat. Test.

Next, we use the ggplot2::ggproto function which allows you to define a new Stat object - which will let us do computation under the hood while building our plot.

Define Stat.

StatMedians <- 
  ggplot2::ggproto(`_class` = "StatMedians",
                   `_inherit` = ggplot2::Stat,
                   compute_group = compute_group_medians,
                   required_aes = c("x", "y"))
NoteYou may have noticed…

  1. … that the naming convention for the ggproto object is written in CamelCase. The new class should also be named the same, i.e. "StatMedians".

  2. … that we inherit from the ‘Stat’ class. In fact, your ggproto object is a subclass – you are inheriting class properties from ggplot2::Stat.

  3. … that the compute_group_medians function is used to define our Stat’s compute_group element. This means that data will be transformed group-wise by our compute definition – i.e. by categories if a categorical variable is mapped.

  4. … that setting required_aes to x and y reflects the compute functions requirements Specifying required_aes in your Stat can improve your user interface. Standard ggplot2 error messages will issue if required aes are not specified, e.g. “stat_medians() requires the following missing aesthetics: x.”

Test Stat.

You can test out your Stat using them in ggplot2 geom_*() functions.

penguins_clean |> 
  ggplot() + 
  aes(x = bill_depth_mm,
      y = bill_length_mm) + 
  geom_point() + 
  geom_point(stat = StatMedians, size = 7) + 
  labs(title = "Testing StatMedians")

NoteYou may have noticed …

… that we don’t use "medians" as the stat argument. But you could! If you prefer, you could write geom_point(stat = "medians", size = 7) which will direct to your new StatMedians under the hood.

Test Stat group-wise behavior

Test group-wise behavior by using a discrete variable with an group-triggering aesthetic like color, fill, or group, or by faceting.

last_plot() + 
  aes(color = species)

You might be thinking, what we’ve done would already be pretty useful to me. Can I just use my Stat as-is within geom_*() functions?

The short answer is ‘yes’! If you just want to use the Stat yourself locally in a script, there might not be much reason to go on to Step 3, user-facing functions. But if you have a wider audience in mind, i.e. internal to organization or open sourcing in a package, probably a more succinct expression of what functionality you deliver will be useful - i.e. write the user-facing functions.

Instead of using a geom_*() function, you might prefer to use the layer() function in your testing step. Occasionally, you must to go this route; for example, geom_vline() contain no stat argument, but you can use the GeomVline in layer(). If you are teaching this content, using layer() may help you better connect this step with the next, defining the user-facing functions.

A test of StatMedians using this method follows. You can see it is a little more verbose, as there is no default for the position argument, and setting the size must be handled with a little more care.

penguins_clean |> 
  ggplot() + 
  aes(x = bill_depth_mm,
      y = bill_length_mm) + 
  geom_point() + 
  layer(geom = GeomPoint, 
        stat = StatMedians, 
        position = "identity", 
        params = list(size = 7)) + 
  labs(title = "Testing StatMedians with layer() function")

Step 3: Define user-facing functions. Test.

In this next section, we define user-facing functions. Doing so is a bit of a mouthful, but see the ‘Pro tip: Use geom_point definition as a template in this step …’ that follows.

Define stat_*() and geom_*() functions.

stat_medians <- function (mapping = NULL, data = NULL,
                          geom = "point", position = "identity",
                          ..., na.rm = FALSE, 
                          show.legend = NA, inherit.aes = TRUE) 
{
    layer(mapping = mapping, data = data, 
          geom = geom, stat = StatMedians, 
          position = position, show.legend = show.legend, 
          inherit.aes = inherit.aes, 
        params = rlang::list2(na.rm = na.rm, ...))
}
NoteYou may have noticed that …

  1. … the stat_*() function name derives from the Stat object’s name, but is snake case. Given naming conventions, a StatBigCircle-based stat_*() function, should be named stat_big_circle().

  2. … StatMedians defines the new layer function and cannot be replaced by the user StatMedians and the computation that defines it will be in effect before the layer is rendered.

  3. … "point" refers to the object GeomPoint and defines the layer’s geom unless otherwise specified.

…

You may be thinking, defining a new stat_*() function is a mouthful that’s probably hard to reproduce from memory. So you might use stat_identity()’s definition as scaffolding to write your own layer. i.e:

  • Type stat_identity in your console to print function contents; copy-paste the function definition.
  • Switch out StatIdentity with your Stat, e.g. StatIndex.
  • Switch out "point" other geom (‘rect’, ‘text’, ‘line’ etc) if needed
  • Final touch, list2 will error without export from rlang, so update to rlang::list2.
stat_identity
function (mapping = NULL, data = NULL, geom = "point", position = "identity", 
    ..., show.legend = NA, inherit.aes = TRUE) 
{
    layer(data = data, mapping = mapping, stat = StatIdentity, 
        geom = geom, position = position, show.legend = show.legend, 
        inherit.aes = inherit.aes, params = list2(na.rm = FALSE, 
            ...))
}
<bytecode: 0x564d30d771c0>
<environment: namespace:ggplot2>

Define geom_*() function

Because users are more accustom to using layers that have the ‘geom’ prefix, you might also define geom with identical properties via aliasing.

geom_medians <- stat_medians

Verbatim aliasing as shown above is a bit of a shortcut and assumes that users will use the ‘geom_*()’ function with the stat-geom combination as-is. (For a discussion, see Constructors in ‘Extending ggplot2: A case Study’ in ggplot2: Elegant Graphics for Data Analysis. This section notes, ‘Most ggplot2 users are accustomed to adding geoms, not stats, when building up a plot.’)

An approach that is more consistent with existing guidance would be to hardcode the Geom and allow the user to change the Stat as follows.

# user-facing function
geom_index <- function(mapping = NULL, data = NULL, 
                         stat = "index", position = "identity", 
                         ..., show.legend = NA, inherit.aes = TRUE) 
{
    layer(data = data, mapping = mapping, stat = stat, 
        geom = GeomPoint, position = position, show.legend = show.legend, 
        inherit.aes = inherit.aes, params = rlang::list2(na.rm = FALSE, 
            ...))
}

However, because it is unexpected to use geom_index() with a Stat other than StatIndex (doing so would remove the index-ness) we think that the verbatim aliasing is a reasonable, time and code saving getting-started approach.

Test/Enjoy your user-facing functions

Test geom_medians()

## Test user-facing.
penguins_clean |>
  ggplot() +
  aes(x = bill_depth_mm, y = bill_length_mm) +
  geom_point() +
  geom_medians(size = 8)  + 
  labs(title = "Testing geom_medians()")

Test group-wise behavior

last_plot() + 
  aes(color = species)

Test stat_*() function with another Geom.

last_plot() + 
  stat_medians(geom = "label", aes(label = species))  + 
  labs(subtitle = "and stat_medians()")

stat_medians <- ggplot2::make_constructor(StatMedians, GeomPoint)
geom_medians <- ggplot2::make_constructor(GeomPoint, StatMedians)
geom_medians_label <- ggplot2::make_constructor(GeomLabel, StatMedians)

# check out the function definitions
geom_medians

penguins_clean |>
  ggplot() +
  aes(x = bill_depth_mm, 
      y = bill_length_mm) +
  geom_point() +
  geom_medians(size = 8)

last_plot() + 
  aes(color = species) 

last_plot() + 
  aes(label = species) +
  geom_medians_label()

Done! Time for a review.

Here is a quick review of the functions and ggproto objects we’ve covered, dropping tests and discussion.

NoteReview 
library(tidyverse)

# Step 1. Define compute
compute_group_medians <- function(data, scales){
  
  data |>
    summarise(x = median(x), y = median(y))
  
}

# Step 2. Define Stat
StatMedians = ggproto(`_class` = "StatMedians",
                      `_inherit` = Stat,
                      required_aes = c("x", "y"),
                      compute_group = compute_group_medians)

# Step 3. Define user-facing functions (user friendly, geom_*() function only shown here)

## use geom_point's definition as a model to follow geom_* conventions: geom is fixed, stat is flexible
geom_medians <- function(mapping = NULL, data = NULL, 
                         stat = "medians", position = "identity", 
                         ..., show.legend = NA, inherit.aes = TRUE) 
{
    layer(data = data, mapping = mapping, stat = stat, 
        geom = GeomPoint, position = position, show.legend = show.legend, 
        inherit.aes = inherit.aes, params = rlang::list2(na.rm = FALSE, 
            ...))
}

Your Turn: write geom_means()

Using the geom_medians Recipe #1 as a reference, try to create a stat_means() function that draws a point at the means of x and y. You may also write convenience geom_*() functions.

Step 00: load libraries, data

Step 0: Use base ggplot2 to get the job done

Step 1: Write compute function. Test.

Step 2: Write Stat.

Step 3: Write user-facing functions

Next up: Recipe 2 geom_id()

How would you write the function which annotates coordinates (x,y) for data points on a scatterplot? Go to Recipe 2.