I want to perform several operations on multiple columns and I can use dplyr::across() to do it that way:
library(tidyverse)
df = tibble(x=1:5, p1=x*2, p2=x*4, p3=x*5)
r1 = df %>%
mutate(across(starts_with("p"), c(inf=~.x-1, sup=~.x+1)))
r1
#> # A tibble: 5 x 10
#> x p1 p2 p3 p1_inf p1_sup p2_inf p2_sup p3_inf p3_sup
#> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 1 2 4 5 1 3 3 5 4 6
#> 2 2 4 8 10 3 5 7 9 9 11
#> 3 3 6 12 15 5 7 11 13 14 16
#> 4 4 8 16 20 7 9 15 17 19 21
#> 5 5 10 20 25 9 11 19 21 24 26
names(r1)
#> [1] "x" "p1" "p2" "p3" "p1_inf" "p1_sup" "p2_inf" "p2_sup"
#> [9] "p3_inf" "p3_sup"
However, this is not very scalable if the function calculates a lot of things as it would be evaluated twice.
Instead, it would be nice if I could use a function that calculates the things that need to be calculated, and then returns a list of the 2 (or more) results.
For instance, consider this example:
#perform heavy calculation on x2 and return 2 flavours of it
f = function(x) {
x2=x #wow, such heavy, very calculate
Sys.sleep(1)
data.frame(inf=x2-10, sup=x2+10)
}
r2 = df %>%
mutate(across(starts_with("p"), f, .names="{.col}_{.fn}"))
r2
#> # A tibble: 5 x 7
#> x p1 p2 p3 p1_1$inf $sup p2_1$inf $sup p3_1$inf $sup
#> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 1 2 4 5 -8 12 -6 14 -5 15
#> 2 2 4 8 10 -6 14 -2 18 0 20
#> 3 3 6 12 15 -4 16 2 22 5 25
#> 4 4 8 16 20 -2 18 6 26 10 30
#> 5 5 10 20 25 0 20 10 30 15 35
names(r2)
#> [1] "x" "p1" "p2" "p3" "p1_1" "p2_1" "p3_1"
map_chr(r2, class)
#> x p1 p2 p3 p1_1 p2_1
#> "integer" "numeric" "numeric" "numeric" "data.frame" "data.frame"
#> p3_1
#> "data.frame"
Created on 2021-10-25 by the reprex package (v2.0.1)
Using rbind() instead of data.frame() would end in the same result with slightly different names (p1_1$inf becomes p1_1[,"inf"]) and a different class (data.frame becomes c("matrix", "array").
Moreover, when using a single function, {.fn} is the position of the function so there are naming issues.
I also tried to unnest() the result but with no success.
Is there a way to get the exact result of my first output using a function in across()?
Since dplyr 1.1.0 (2023-01-29), across() has a .unpack argument that you only need to set to TRUE for this to work.
Actually, this has been considered in an issue on the Github of dplyr: https://github.com/tidyverse/dplyr/issues/5563#issuecomment-721769342.
There, @romainfrancois gives an incredibly useful solution as this unpackross() function:
library(tidyverse)
f = function(x) tibble(inf=x-10, sup=x+10)
unpackross = function(...) {
out = across(...)
tidyr::unpack(out, names(out), names_sep = "_")
}
df = tibble(x=1:5, p1=x*2, p2=x*4, p3=x*5)
r2 = df %>%
mutate(unpackross(starts_with("p"), f, .names="{.col}_{.fn}"))
r2
#> # A tibble: 5 x 10
#> x p1 p2 p3 p1_1_inf p1_1_sup p2_1_inf p2_1_sup p3_1_inf p3_1_sup
#> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 1 2 4 5 -8 12 -6 14 -5 15
#> 2 2 4 8 10 -6 14 -2 18 0 20
#> 3 3 6 12 15 -4 16 2 22 5 25
#> 4 4 8 16 20 -2 18 6 26 10 30
#> 5 5 10 20 25 0 20 10 30 15 35
names(r2)
#> [1] "x" "p1" "p2" "p3" "p1_1_inf" "p1_1_sup"
#> [7] "p2_1_inf" "p2_1_sup" "p3_1_inf" "p3_1_sup"
map_chr(r2, class)
#> x p1 p2 p3 p1_1_inf p1_1_sup p2_1_inf p2_1_sup
#> "integer" "numeric" "numeric" "numeric" "numeric" "numeric" "numeric" "numeric"
#> p3_1_inf p3_1_sup
#> "numeric" "numeric"
Created on 2021-10-26 by the reprex package (v2.0.1)
Hopefully, there will be an unpack parameter in across() one day! (add a +1 on my suggestion here if you agree)