How to use dynamic tidy-select expressions in dplyr::select()?

Question

I need to select variables dynamically on multiple expressions. Consider the following example:

library(tidyverse)
set.seed(42)

df <- tibble(
  grey_dog = runif(n = 69),
  white_bear = runif(n = 69),
  blue_oyster = runif(n = 69),
  white_lobster = runif(n = 69),
  green_dog = runif(n = 69)
)

df %>% 
  dplyr::select(
    (contains("dog") & contains("green")) | 
    (contains("white") & contains("bear"))
  )

Instead of explicitly selecting, I have vectors containing the information I want to base my selection on:

x <- c(green = "dog", white = "bear")

So I was hoping to concatenate a string that can be used as a tidy-select:

s <- paste0("(", paste0("contains(", names(x),") & contains(", x, ")"), ")", collapse = " | ")
dplyr::select(df, s)

This fails with:

Error: Can't subset columns that don't exist.
x Column `(contains(green) & contains(dog)) | (contains(white) & contains(bear))` doesn't exist.
Run `rlang::last_error()` to see where the error occurred.

Any ideas on how to accomplish this?

score 4 · Accepted Answer · answered Feb 25 '22 at 19:43

You can't just pass in a string. A string is not the same as an expression. One way is to use purrr and rlang to build the expression and then inject that into the select

library(purrr)
library(rlang)
query <- map2(
  map(x, ~expr(contains(!!.x))),
  map(names(x), ~expr(contains(!!.x))),
  ~expr((!!.x & !!.y))) %>% 
reduce(~expr(!!.x | !!.y))
dplyr::select(df, !!query)

Though if you really wanted to build the code as a string, then you would just need to parse that string into an expression first using rlang::parse_expr. You just need to add some quotes to your string so it exactly matches the code you used before

s <- paste0("(", paste0("contains(\"", names(x),"\") & contains(\"", x, "\")"), ")", collapse = " | ")
dplyr::select(df, !!rlang::parse_expr(s))

Mikko Marttila · Answer 2 · 2022-02-25T20:30:20.423

@MrFlick's answer is the way to go for building a tidyselect expression.

However, since you’re building the selection programmatically already: instead of a tidyselect expression, it might be easier to just do the corresponding computations on the names yourself.

library(purrr)
set.seed(42)

df <- data.frame(
  grey_dog = runif(n = 5),
  white_bear = runif(n = 5),
  blue_oyster = runif(n = 5),
  white_lobster = runif(n = 5),
  green_dog = runif(n = 5)
)

x <- list(
  c("green", "dog"),
  c("white", "bear")
)

sel <- x |>
  map(sapply, grepl, names(df)) |>
  map(apply, 1, all) |>
  reduce(`|`)

df[, sel]
#>   white_bear  green_dog
#> 1  0.5190959 0.90403139
#> 2  0.7365883 0.13871017
#> 3  0.1346666 0.98889173
#> 4  0.6569923 0.94666823
#> 5  0.7050648 0.08243756

Or, you could apply the helpers inside the selecting function with map() and then do the set operatins corresponding to & and | in a similar sequence of steps:

df |>
  dplyr::select(
    x |>
      map(map, contains) |>
      map(reduce, intersect) |>
      reduce(union)
  )
#>    green_dog white_bear
#> 1 0.90403139  0.5190959
#> 2 0.13871017  0.7365883
#> 3 0.98889173  0.1346666
#> 4 0.94666823  0.6569923
#> 5 0.08243756  0.7050648

How to use dynamic tidy-select expressions in dplyr::select()?

2 Answers2