across(), but over two sets of vector

[abichat]

Sometimes, when doing data wrangling, you need to apply functions not only on multiple columns, but also on multiple pairs of columns. This is what the {dplyover} package allows with its across2() and across2x() functions:

# remotes::install_github("TimTeaFan/dplyover")
library(dplyr)
library(dplyover)

iris %>% 
  group_by(Species) %>%
  summarise(across2x(starts_with("Sepal"), starts_with("Petal"), cor))
#> # A tibble: 3 × 5
#>   Species   Sepal.Length_Petal.L…¹ Sepal.Length_Petal.W…² Sepal.Width_Petal.Le…³
#>   <fct>                      <dbl>                  <dbl>                  <dbl>
#> 1 setosa                     0.267                  0.278                  0.178
#> 2 versicol…                  0.754                  0.546                  0.561
#> 3 virginica                  0.864                  0.281                  0.401
#> # ℹ abbreviated names: ¹​Sepal.Length_Petal.Length, ²​Sepal.Length_Petal.Width,
#> #   ³​Sepal.Width_Petal.Length
#> # ℹ 1 more variable: Sepal.Width_Petal.Width <dbl>

cor(iris[1:50,]$Sepal.Length, iris[1:50,]$Petal.Length)
#> [1] 0.2671758

^{Created on 2024-07-17 with reprex v2.1.0}

The creation of the {dplyover} began with this issue, which –at that time– recommended putting these features in an adjacent package.

{dplyover} has really saved me and my colleagues a huge amount of time. However, it is not on CRAN, some bugs [1, 2] have been found, enhancement in the {dplyr} package are not currently supported, and the last commit is 3 years old.

Do you think these really useful functions could be included in the {dplyr} package? Thanks!

[ggrothendieck]

It can be done with dplyr/purrr like this but it certainly is a lot easier and more obvious with across2x

library(dplyr)
library(purrr)

iris %>%
  reframe(map(pick(contains("Sepal")), \(x) map(pick(contains("Petal")), \(y) cor(x, y))) %>%
      list_flatten %>%
      as_tibble,
    .by = Species
  )
##      Species Sepal.Length_Petal.Length Sepal.Length_Petal.Width Sepal.Width_Petal.Length Sepal.Width_Petal.Width
## 1     setosa                 0.2671758                0.2780984                0.1777000               0.2327520
## 2 versicolor                 0.7540490                0.5464611                0.5605221               0.6639987
## 3  virginica                 0.8642247                0.2811077                0.4010446               0.5377280

If it is ok to use the fact that cor itself can do crossing then we can avoid the nested map calls although still much less obvious than across2x

library(dplyr)
library(tidyr)

iris %>%
  reframe(cor(pick(contains("Sepal")), pick(contains("Petal"))) %>%
       as.data.frame.table(responseName = "value") %>%
       pivot_wider(names_from = 1:2),
    .by = Species)
##      Species Sepal.Length_Petal.Length Sepal.Width_Petal.Length Sepal.Length_Petal.Width Sepal.Width_Petal.Width
## 1     setosa                 0.2671758                0.1777000                0.2780984               0.2327520
## 2 versicolor                 0.7540490                0.5605221                0.5464611               0.6639987
## 3  virginica                 0.8642247                0.4010446                0.2811077               0.5377280