dplyr 1.0.0: working across columns

  dplyr, dplyr-1-0-0

  Hadley Wickham

Today, I wanted to talk a little bit about the new across() function that makes it easy to perform the same operation on multiple columns.


Update: as of June 1, dplyr 1.0.0 is now available on CRAN! Read all about it or install it now with install.packages("dplyr").


Motivation

It’s often useful to perform the same operation on multiple columns, but copying and pasting is both tedious and error prone:

df %>% 
  group_by(g1, g2) %>% 
  summarise(a = mean(a), b = mean(b), c = mean(c), d = mean(c))

You can now rewrite such code using across(), which lets you apply a transformation to multiple variables selected with the same syntax as select() and rename():

df %>% 
  group_by(g1, g2) %>% 
  summarise(across(a:d, mean))

# or with a function
df %>% 
  group_by(g1, g2) %>% 
  summarise(across(where(is.numeric), mean))

You might be familiar with summarise_if() and summarise_at() which we previously recommended for this sort of operation. Later in the blog post we’ll come back to why we now prefer across(). But for now, let’s dive into the basics of across().

Basic usage

across() has two primary arguments:

  • The first argument, .cols, selects the columns you want to operate on. It uses the tidy select syntax so you can pick columns by position, name, function of name, type, or any combination thereof using Boolean operators.

  • The second argument, .fns, is a function or list of functions to apply to each column. You can use also purrr style formulas like ~ .x / 2.

Here are a couple of examples of across() used with summarise():

library(dplyr, warn.conflicts = FALSE)

starwars %>% 
  summarise(across(where(is.character), n_distinct))
#> # A tibble: 1 x 8
#>    name hair_color skin_color eye_color   sex gender homeworld species
#>   <int>      <int>      <int>     <int> <int>  <int>     <int>   <int>
#> 1    87         13         31        15     5      3        49      38

starwars %>% 
  group_by(species) %>% 
  filter(n() > 1) %>% 
  summarise(across(c(sex, gender, homeworld), n_distinct))
#> `summarise()` ungrouping output (override with `.groups` argument)
#> # A tibble: 9 x 4
#>   species    sex gender homeworld
#>   <chr>    <int>  <int>     <int>
#> 1 Droid        1      2         3
#> 2 Gungan       1      1         1
#> 3 Human        2      2        16
#> 4 Kaminoan     2      2         1
#> 5 Mirialan     1      1         1
#> 6 Twi'lek      2      2         1
#> 7 Wookiee      1      1         1
#> 8 Zabrak       1      1         2
#> 9 <NA>         1      1         3

starwars %>% 
  group_by(homeworld) %>% 
  filter(n() > 1) %>% 
  summarise(across(where(is.numeric), mean, na.rm = TRUE), n = n())
#> `summarise()` ungrouping output (override with `.groups` argument)
#> # A tibble: 10 x 5
#>    homeworld height  mass birth_year     n
#>    <chr>      <dbl> <dbl>      <dbl> <int>
#>  1 Alderaan    176.  64         43       3
#>  2 Corellia    175   78.5       25       2
#>  3 Coruscant   174.  50         91       3
#>  4 Kamino      208.  83.1       31.5     3
#>  5 Kashyyyk    231  124        200       2
#>  6 Mirial      168   53.1       49       2
#>  7 Naboo       175.  64.2       55      11
#>  8 Ryloth      179   55         48       2
#>  9 Tatooine    170.  85.4       54.6    10
#> 10 <NA>        139.  82        334.     10

Other cool features

You’ll find a lot more about across() in vignette("colwise"). There are three cool features you might be particularly interested in:

Why across()?

If you’ve tackled this problem with an older version of dplyr, you might’ve used one of the functions with an _if, _at, or _all suffix. These functions solved a pressing need and are used by many people, but are now superseded. This means that they’ll stay around, but will only receive critical bug fixes.

Why did we decide to move away from these functions in favour of across()?

  1. across() makes it possible to compute useful summaries that were previously impossible. For example, it’s now easy to summarise numeric vectors with one function, factors with another, and still compute the number of rows in each group:

    df %>%
      group_by(g1, g2) %>% 
      summarise(
        across(where(is.numeric), mean), 
        across(where(is.factor), nlevels),
        n = n(), 
      )
    
  2. across() reduces the number of functions that dplyr needs to provide. This makes dplyr easier for you to use (because there are fewer functions to remember) and easier for us to develop (since we only need to implement one function for each new verb, not four).

  3. With the where() helper, across() unifies _if and _at semantics, allowing combinations that used to be impossible. For example, you can now transform all numeric columns whose name begins with “x”: across(where(is.numeric) & starts_with("x")).

  4. across() doesn’t need vars(). The _at() functions are the only place in dplyr where you have to use vars(), which makes them unusual, and hence harder to learn and remember.

Why did it take it long to discover across()? Surprisingly, the key idea that makes across() works came out of our low-level work on the vctrs package, where we learnt that you can have a column of a data frame that is itself a data frame. It’s a bummer that we had a few false starts before we discovered across(), but even with hindsight, I don’t see how we could’ve skipped the intermediate steps.

Converting existing code

If you want to update your existing code to use across() instead of the _if, _at, or _all() functions, it’s generally straightforward:

  • Strip the _if(), _at() and _all() suffix off the function.

  • Call across(). The first argument will be:

    1. For _if(), the old second argument.
    2. For _at(), the old second argument. If there was a single element in vars() you can remove vars(), otherwise replace it with c().
    3. For _all(), everything().

    The subsequent arguments can be copied as is.

Here are a few examples of this process:

df %>% mutate_if(is.numeric, mean, na.rm = TRUE)
# ->
df %>% mutate(across(where(is.numeric), mean, na.rm = TRUE))

df %>% mutate_at(vars(x, starts_with("y")), mean, na.rm = TRUE)
# ->
df %>% mutate(across(c(x, starts_with("y")), mean, na.rm = TRUE))

df %>% mutate_all(mean, na.rm = TRUE)
# ->
df %>% mutate(across(everything(), mean, na.rm = TRUE))

If you’ve used multiple _if/_at/_all functions in a row, you should also consider if it’s now possible to collapse them into a single call, using the new features of across().

Again, you don’t need to worry about these functions going away in the short-term, but it’s good practice to keep your code up-to-date. Note, however, that across() currently has a little more overhead than the older approaches so it will be a little slower. We have a plan to improve the performance in dplyr 1.1.0.