Equivalent of `break` in purrr::map

Question

Say I want to run a loop until a condition is met, at which point the result is saved and the loop exits:

library(tidyverse)

for (i in 1:5) {

  df <- iris %>% select(i) %>% head(2)

  if (names(df) == "Petal.Width") {
    out <- df
    break 

  }
}

out

How can I rewrite this using purr::map without having to evaluate each i?

Doing the following gives the result I need, but has to evaluate 5 times, whereas the for loop only 3 times:

fun <- function(x) {

  df <- iris %>% select(x) %>% head(2)

  if (names(df) == "Petal.Width") {
  return(df)
  }
}

map_df(1:5, fun)

Answer 1

There is no equivalent. In fact, one thing that makes map (and similar functions) so superior to general loops in terms of readability is that they have absolutely predicable behaviour: they will execute the function exactly once for each element, no exceptions (except, uh, if there’s an exception: you could raise a condition via stop to short-circuit execution, but this is very rarely advisable).

Instead, your case doesn’t call for map, it calls for something along the lines of purrr::keep or purrr::reduce.

Think of it this way: map, reduce, etc. are abstractions which correspond to specific special cases of the more general for loop. Their purpose is to make clear which special case is being handled. As a programmer, your task then becomes to find the right abstraction.

In your particular case I would probably completely rewrite the statement using dplyr so giving a “best” purrr solution is hard: the best solution is not to use purrr. That said, you could use purrr::detect as follows:

names(iris) %>%
    detect(`==`, 'Sepal.Width') %>%
    `[`(iris, .) %>%
    head(2)

Or

seq_along(iris) %>%
    detect(~ names(iris[.x]) == 'Sepal.Width') %>%
    `[`(iris, .) %>%
    head(2)

… but really, here’s dplyr for comparison:

iris %>%
    select(Sepal.Width) %>%
    head(2)

Answer 2

1) callCC can be used to get this effect:

callCC(function(k) {
  fun2 <- function(x) {
    print(x) # just to show that x = 5 is never run
    df <- iris %>% select(x) %>% head(2)
    if (names(df) == "Petal.Width") k(df)
  }
  map_df(1:5, fun2)
})

giving:

[1] 1
[1] 2
[1] 3
[1] 4
  Petal.Width
1         0.2
2         0.2

1a) If it is important to use fun without change then try this instead:

callCC(function(k) map_df(1:5, ~ if (!is.null(df <- fun(.x))) k(df)))

2) purrr::reduce An alternative is to use reduce from purrr (or Reduce from base R):

f <- function(x, y) if (is.null(x)) fun(y) else x
reduce(1:5, f, .init = NULL)

This is not as good as (1) and (1a) from the viewpoint that it will still involve iterating over each element of 1:5 but will only invoke fun for 1:4. In contrast (1) and (1a) actually return after running fun or fun2 on 4.