purrr map equivalent of nested for loop

Question

What is the purrr::map equivalent of:

for (i in 1:4) {
  for (j in 1:6) {
    print(paste(i, j, sep = "-"))
  }
}

OR

lapply(1:4, function(i) 
  lapply(1:6, function(j) 
    print(paste(i, j, sep = "-"))))

Conceptually, what I'm not getting is how to refer to the outer loop in the inner map function.

map(1:4, ~ map(1:6, ~ print(paste(.x, ????, sep = "-")))

`outer(1:4, 1:6, paste, sep = '-')` gives a nice matrix. `cross2(1:4, 1:6) %>% map_chr(paste, collapse = '-')` gives a character vector. — alistaire, Feb 18 '18 at 02:28
or in base for a character vector, `do.call(paste, c(expand.grid(1:4, 1:6), sep = '-'))` — alistaire, Feb 18 '18 at 02:36
I find it improves readability to use `%>%` between consecutive `purrr` map-like functions. E.i pass the output list of one map as the input list to another map function. — Kresten, Apr 11 '19 at 09:27

r2evans · Answer 1 · 2018-02-18T06:14:34.013

47

The use of function formulas (~) is a little limited when trying to nest like this, since it is perfectly unclear which level of map you are attempting to reference. (Well, that's not correct. It's perfectly clear to me that it is referencing inside-out, and since they both use the same nomenclature, the outer variables are being masked by the inner variables.)

I think your best way around it is to not use the formula method, instead using immediate/anonymous (or predefined) functions:

library(purrr)
str(map(1:2, function(x) map(1:3, function(y) paste(x, y, sep = "-"))))
# List of 2
#  $ :List of 3
#   ..$ : chr "1-1"
#   ..$ : chr "1-2"
#   ..$ : chr "1-3"
#  $ :List of 3
#   ..$ : chr "2-1"
#   ..$ : chr "2-2"
#   ..$ : chr "2-3"

edited Feb 18 '18 at 06:14

answered Feb 18 '18 at 02:09

r2evans

141,215
6
77
149

3

You could also quote one and pass the function name and arguments separately for the other: `map(1:4, ~map_chr(1:6, paste, .x, sep = '-'))` – alistaire Feb 18 '18 at 02:42
Good point, works much better when one of the levels can be literal. Thanks! – r2evans Feb 18 '18 at 02:45

moodymudskipper · Accepted Answer · 2018-12-02T15:43:50.100

As @r2evans points out, the .x from your first call is masked. however you can create a lambda function that takes 2 parameters .x and .y, and assign the previous .x to the new .y through the ... argument.

I'll use walk rather than map as in this case you're only interested in side effects (printing)

walk(1:4,~ walk(1:6, ~ print(paste(.x, .y, sep = "-")),.y=.x))

Another option is to use expand.grid to lay out the combinations, and then iterate on those with pwalk (or pmap in other circumstances)

purrr::pwalk(expand.grid(1:4,1:6),~print(paste(.x, .y, sep = "-")))

Output in both cases:

[1] "1-1"
[1] "2-1"
[1] "3-1"
[1] "4-1"
[1] "5-1"
[1] "6-1"
[1] "1-2"
[1] "2-2"
[1] "3-2"
[1] "4-2"
[1] "5-2"
[1] "6-2"
[1] "1-3"
[1] "2-3"
[1] "3-3"
[1] "4-3"
[1] "5-3"
[1] "6-3"
[1] "1-4"
[1] "2-4"
[1] "3-4"
[1] "4-4"
[1] "5-4"
[1] "6-4"

score 5 · Answer 3 · answered Jun 24 '19 at 18:04

5

Just Running through this now.

walk(1:4,~ walk(1:6, ~ print(paste(.x, .y, sep = "-")),.y=.x)) 
[1] "1-1"
[1] "2-1"
[1] "3-1"
[1] "4-1"
[1] "5-1"
[1] "6-1"
[1] "1-2"

and

purrr::pwalk(expand.grid(1:4,1:6),~print(paste(.x, .y, sep = "-")))
[1] "1-1"
[1] "2-1"
[1] "3-1"
[1] "4-1"
[1] "1-2"

but to match your nested for loops exactly it fiddled and this works.

for (i in 1:4) {
  for (j in 1:6) {
    print(paste(i, j, sep = "-"))
  }
}
[1] "1-1"
[1] "1-2"
[1] "1-3"
[1] "1-4"
[1] "1-5"
[1] "1-6"
[1] "2-1"

purrr::pwalk(expand.grid(1:6,1:4),~print(paste(.y, .x, sep = "-")))
[1] "1-1"
[1] "1-2"
[1] "1-3"
[1] "1-4"
[1] "1-5"
[1] "1-6"
[1] "2-1"

#or even a map of this
walk(1:4,~ walk(1:6, ~ print(paste(.y, .x, sep = "-")),.y=.x))

I have yet to figure out why the .y=.x is at the end though.

answered Jun 24 '19 at 18:04

Shadowhawk

319
1
3
9

I think it might be a fluke that it works. My speculation is that creating a function with `~` looks for possible recognized arguments (`.`, `.x`, `.y`, `..1`, etc.) and the last argument (`.y=.x`) is somehow aliasing what was `.x` in the outer walk is being set to `.y` in the inner walk. You can replace `.y` with `..1`, but not with other variables. – Brian Diggs Apr 14 '20 at 21:56
Using the formula notation is like creating a function of form `function (..., .x = ..1, .y = ..2, . = ..1) {}`, so you can pass an arg to `.y` through the `walk` call, and you can pass it the `.x` of the outer walk call since it's not overriden by anything at this point, not a fluke! – moodymudskipper Oct 26 '20 at 22:56

Jordan Mandel · Answer 4 · 2020-08-25T05:23:24.743

Here is an addition to the already very good answers and answer-comments. I wanted to make a single purr-like function that accomplishes the OP's goals. So I made a loop_map function that behaves analogously to the main Purrr map functions.

loop_map <- function(why, ecks, fun) {
  
  # 2: for every call of this (the previous .f) the new .f is called for each
  # value of ecks, supplied the same value of why each time
  iterate_over_x = function(x_in,y_in,fun_in){
    return(pmap(.l = list(x = x_in), .f = fun_in ,y = y_in ) %>%
 set_names(nm = as.character(x_in))) 
  }
  
  # 1: this ".f"  argument is called once for each element of why, and is 
  # supplied one value of why and every value of ecks each time
  pmap(.l = list(y_in = why), .f = iterate_over_x, x_in = ecks, fun_in = fun) %>% 
set_names(nm = as.character(why))
  
}



my_paste <- function(x,y) {
  paste(x,y)
}



loop_map(list("a","b"),list("c","d"),my_paste)

As a bonus I named the output so that one can index it more easily, or somehow convert it to a dataframe. I would like to improve this function by adding capabilities to loop over arbitrarily many input lists, and possibly to use functions that take ... arguments (right now everything has to be named). If anyone has an idea for how to do this feel free to let me know.

purrr map equivalent of nested for loop

4 Answers4

Linked

Related