Writing functions in R

R
intermediate
Published

2024-07-09

Previous attendees have said…

  • 17 previous attendees have left feedback
  • 100% would recommend this session to a colleague
  • 88% said that this session was pitched correctly

NoteThree random comments from previous attendees
  • Good session, as a R beginner learned some interesting things about how to write functions
  • Settings for input parameters, how to include more general inputs, and the possible pitfalls e.g. non-vectorised elements, were helpful
  • Presentation was very clear and I was able to check up on an unclear area for me by checking out the slides afterwards. Very useful to see curly brackets & vectorize()

Welcome!

  • this session is an 🌶🌶 intermediate practical designed for those with some R experience

Session outline

  • why functions?
  • basic syntax
  • understanding how functions work
  • vectorised functions
  • the mysteries of the paired curly brackets {{}} and the dots ...
  • anonymous functions
  • a bunch of ways to call functions

Why functions?

  • most beginners write repetitious code
  • repetitious code is hard to maintain
  • functions give you an easy way of repeating chunks of code

Basic syntax

  • think of this as a way of repeating yourself
  • in time-honoured fashion…
hi_wrld <- function(){
  "hello world"
}

hi_wrld()
[1] "hello world"

Adding arguments

  • most of the time, you’ll want to add arguments to your function
    • add a variable name inside the round bracket of function
    • use that variable name in your function body
hi_wrld_n <- function(n){
  paste(rep("hello world", n))
}

hi_wrld_n(4)
[1] "hello world" "hello world" "hello world" "hello world"

Another argument

  • you can add another argument
  • either position or name can be used in the function call
hi_name_n <- function(name, n){
  rep(paste("hello", name) , n)
}

hi_name_n("sue", 4)
[1] "hello sue" "hello sue" "hello sue" "hello sue"
hi_name_n(n = 3, name = "tango") # evil but legal
[1] "hello tango" "hello tango" "hello tango"

Defaults

hi_name_n_def <- function(n, name = "janelle"){
  rep(paste("hello", name) , n)
}

hi_name_n_def(n = 4)
[1] "hello janelle" "hello janelle" "hello janelle" "hello janelle"
hi_name_n_def(n = 2, name = "bruce")
[1] "hello bruce" "hello bruce"

Namespacing

If you plan to use non-base R functions in your function, it’s always a good idea to namespace them:

hp_av <- function(cyls = 4){
  mtcars |>
    dplyr::filter(cyl == cyls) |>
    dplyr::summarise(mean_hp = round(mean(hp)))
}

hp_av(8)
  mean_hp
1     209

Thinking about functions

  • R functions are ordinary R objects, usually bound to a function name
  • there are several ways of investigating functions using ordinary R code. Two ways of investigating aspects of functions we’ve already discussed:
hp_av |>
  formals() # the arguments
$cyls
[1] 4
hp_av |>
  body() # the body
{
    dplyr::summarise(dplyr::filter(mtcars, cyl == cyls), mean_hp = round(mean(hp)))
}

Operators and functions

  • most R operators are really functions too - properly referred to as infix functions. You can use them in the standard prefix way using backticks:
`+`(2, 2)
[1] 4

If you’re feeling …experimental, you can also convert a standard function to an infix function by wrapping the function name in %%:

`%tri%` <- function(a, b){
  sqrt(a^2 + b^2)
}

3 %tri% 4
[1] 5

class()/typeof()

For user-defined functions, and functions from packages, this means class() gives sensible results:

class(hp_av)
[1] "function"
typeof(hp_av) # run away!
[1] "closure"

Primitive functions

Note that’s not true for some built-in functions written in C:

typeof(sum)
[1] "builtin"

Environments

hp_av |>
  environment() # the environment
<environment: R_GlobalEnv>
  • Think of the environment as the space in which the function is executed
  • Much more to say really - but Advanced R is the right place for that

Scoping

  • the reason environments are important: names defined within scope mask those outwith:
a <- 10

scope_test <- function(n) {
  a <- 5
  n + a
}

scope_test(5)
[1] 10

Name masking

  • if names aren’t defined within the environment of the function, it’ll look one level up
a <- 10

scope_test2 <- function(n) {
  # a <- 5
  n + a
}

scope_test2(5)
[1] 15

Lazy evaluation

  • names get used at execution only. This explains the following odd behaviour:
a <- 10

scope_test3 <- function(n) {
  # a <- 5
  a
}

scope_test3()
[1] 10

You’d usually expect the absence of n in the function call to produce an error. But because the body never uses n, no problem!

Vectorised functions

round(c(1.2, 3.2, 5.4, 2.7), 0)
[1] 1 3 5 3
  • that means that mostly, our functions will end up vectorised without us doing any work at all
div_seven_n_round <- function(nums){
  round(nums / 7, 0)
}

numbers <- rnorm(4, 5, 50)

numbers
[1] -51.61748 -46.81730  52.14717  12.21416
div_seven_n_round(numbers)
[1] -7 -7  7  2
  • but there are a few cases where that can fail: most famously, using if/else 
is_even <- function(n){
  
  if(n %% 2){
  paste(n, "is odd")
} else {
  paste(n, "is even")
}
  
}
is_even(9)
[1] "9 is odd"
is_even(10)
[1] "10 is even"
try(is_even(9:10))
Error in if (n%%2) { : the condition has length > 1

Three solutions

vectorize with Vectorize 

is_even_v <- Vectorize(is_even)
is_even_v(9:10)
[1] "9 is odd"   "10 is even"

apply

lapply(9:10, is_even)
[[1]]
[1] "9 is odd"

[[2]]
[1] "10 is even"
purrr::map(9:10, is_even)
[[1]]
[1] "9 is odd"

[[2]]
[1] "10 is even"

refactor

  • refactor to avoid scalar functions
is_even_rf <- function(n){
  ifelse(n %% 2, paste(n, "is odd"), paste(n, "is even"))
}
is_even_rf(9:10)
[1] "9 is odd"   "10 is even"

{{}}

What’s the problem?

mtcars |>
  dplyr::summarise(average = round(mean(hp)))
  average
1     147

so

carmo <- function(column){
  mtcars |>
    dplyr::summarise(average = round(mean(column)))
  }
  • but…
try(carmo(hp))
Error in dplyr::summarise(mtcars, average = round(mean(column))) : 
  ℹ In argument: `average = round(mean(column))`.
Caused by error:
! object 'hp' not found

object ‘hp’ not found

  • we get used to R (and particularly tidyverse) helping us with some sugar when selecting column by their names
    • mtcars$hp / mtcars |> select(hp)
    • effectively, we’re just able to specify hp like an object, and R figures out the scope etc for us
  • that misfires inside functions. R isn’t sure where to look for an object called hp

Enter {{}} 

carmo_woo <- function(column){
  mtcars |>
    dplyr::summarise(average = round(mean({{column}})))
}

carmo_woo(hp)
  average
1     147
  • for 95% of purposes, take {{}} as a purely empirical fix
  • but, if you’re very enthusiastic:

…

  • pass arbitrary arguments into/through a function with …
dotty <- function(n, ...){
  rep(paste(..., collapse = ""), n)
}

dotty(4, letters[1:5])
[1] "abcde" "abcde" "abcde" "abcde"

If you need to collect values from ..., rlang::list2() is probably the right way to do that:

list_out <- function(...){
  rlang::list2(...)
}

list_out(1, 4, 9)
[[1]]
[1] 1

[[2]]
[1] 4

[[3]]
[1] 9

Anonymous functions

In some cases, you might find yourself creating a function that’s only going to be used in a single location. In that case, it’s possible to define a nameless (anonymous) function, which is concise-but-nasty. For example:

(\(x) x * 2)(5)
[1] 10

Here:

  • \(x) defines an anonymous function that takes one argument x.
  • x * 2 is the body of the function which determines how the function works
  • (5) is the value that we’re supplying to the function

Most usually, that anonymous function won’t be used with a single value, but within a pipe. Here, the supplied value is dropped, and substituted by the values passing along the pipe:

mtcars |>
  dplyr::select(wt) |>
  (\(x) x * 2)() |> # Double the weight of all the values
  dplyr::slice_max(wt, n = 3)
                        wt
Lincoln Continental 10.848
Chrysler Imperial   10.690
Cadillac Fleetwood  10.500

Anonymous functions and the magrittr pipe

The magrittr pipe (%>%) offers a simple method of avoiding the horrid anonymous function syntax using the . placeholder:

library(magrittr)

mtcars %>%
  dplyr::select(wt) %>%
  { . * 2 } %>% # Double the weight of all the values
  dplyr::slice_max(wt, n = 3)
                        wt
Lincoln Continental 10.848
Chrysler Imperial   10.690
Cadillac Fleetwood  10.500

Calling functions from lists

funs <- list(
  half = function(x) x / 2,
  double = function(x) x * 2
)

funs$half(89)
[1] 44.5

do.call()

  • For when you want to call a function over a stored set of arguments - e.g. where you have values held as a list:
dfs <- list(
  a = tibble::tibble(x = 1, y = 2),
  b = tibble::tibble(x = 3, y = 4)
)

do.call(rbind, dfs)
# A tibble: 2 × 2
      x     y
* <dbl> <dbl>
1     1     2
2     3     4

Resources

  • best = home made! Refactor something simple in your code today.
  • hard to beat the treatment of functions in R4DS
  • the Rlang page on data masking is surprisingly sane for such a complicated area