flatten_names()

[lionel-]

To provide a way of combining names prior to flattening:

x <- list(A = list(foo = 1, bar = 2), B = list(baz = 3)

x %>%
  flatten_names(~paste(.x, .y, sep = " :: ")) %>%
  flatten()
#> $`A :: foo`
#> [1] 1
#>
#> $`A :: bar`
#> [1] 2
#>
#> $`B :: baz`
#> [1] 3

The definition would be

flatten_names <- function(.x, .f = paste, ...) { ... }

So this would work as well:

x %>% flatten_names(sep = " :: ")

[hadley]

Or should this be an argument to flatten()? (Like .name_repair in tibble())

[ColinFay]

Hey,

I was about to open an issue after needing something related, but I can see it's already been thought about :) Feel free to tell if this is not related, I can open another issue.

I think an extra argument to flatten would be nice (maybe keep_names and sep ?)

Here is a minimal reprex of my use case :

I'm requesting an API, which return nested JSON. Let's say the result is:

{
  "object": {
    "colors": {
      "blue": 1,
      "green": 2,
      "brown": 3
    }, 
    "animals": {
      "cat": 4,
      "mouse": 4,
      "dog": 6
    }
  }
}

Which would be the R equivalent of

res <- list(
  object = list(
    colors = list(
      blue = 1, 
      green = 2,
      brown = 3
    ), 
    animals = list(
      cat = 4, 
      mouse = 5,
      dog = 6
    )
  )
)

I want to turn this into a data.frame, but I want to keep the information that "blue" is in the "color" sublist (in my case it has meaning to keep the middle names), and this is lost when using flatten :

map_df(res, flatten_df)
# A tibble: 2 x 6
   blue green brown   cat mouse   dog
  <int> <int> <int> <int> <int> <int>
1     1     2     3     4     4     6

While what I would want is to have something like

flatten_df(res[[1]], keep_names = TRUE, sep = "_") %>% names()
[1] "colors_blue"  "colors_green" "colors_brown" "animals_cat"   "animals_mouse" "animals_dog"

So now I'm handling is with :

map_df(res, ~ tibble(
  colors_blue = .x$colors$blue, 
  colors_green = .x$colors$green, 
  colors_brown = .x$colors$brown, 
  animals_cat = .x$animals$cat, 
  animals_mouse = .x$animals$mouse, 
  animals_dog = .x$animals$dog, 
))

But it's unefficient and would fail if at some point the JSON structure changes.

And this would also make sense on this kind of cases :

l <- list(
  a = list(
    b = 12
  ), 
  c = list(
    b = 13
  )
)

flatten(l)
$b
[1] 12

$b
[1] 13

[ColinFay]

I don't know if you would prefer a C implementation or handling this in R, but here is a suggestion :

# Calling devtools loal all so I can .Call
devtools::load_all("~/Seafile/documents_colin/R/purrr/")
#> Loading purrr
flatten <- function(.x, keep_names = FALSE, sep = "_") {
  
  # Handling unnamed list & keep_names = TRUE
  if (keep_names & all(is.null(names(.x)))) {
    abort("You can't keep names of an unnamed list")
  }
  
  res <- .Call(flatten_impl, .x)
  
  if (keep_names)  {
    nm <- paste(names(.x), map_chr(.x, names), sep = sep)
    return(set_names(res, nm))
  }
  res
}


l <- list(
  a = list(
    b = 12
  ), 
  c = list(
    b = 13
  )
)

flatten(l)
#> $b
#> [1] 12
#> 
#> $b
#> [1] 13
flatten(l, keep_names = TRUE)
#> $a_b
#> [1] 12
#> 
#> $c_b
#> [1] 13

flatten(l, keep_names = TRUE, sep = "@")
#> $`a@b`
#> [1] 12
#> 
#> $`c@b`
#> [1] 13

x <- rerun(2, sample(4))

flatten(x)
#> [[1]]
#> [1] 2
#> 
#> [[2]]
#> [1] 3
#> 
#> [[3]]
#> [1] 4
#> 
#> [[4]]
#> [1] 1
#> 
#> [[5]]
#> [1] 4
#> 
#> [[6]]
#> [1] 2
#> 
#> [[7]]
#> [1] 3
#> 
#> [[8]]
#> [1] 1
flatten(x, keep_names = TRUE)
#> Error: You can't keep names of an unnamed list

(I removed the backtrack from the output)

^{Created on 2019-01-10 by the reprex package (v0.2.1)}

Would be happy to work on this and PR if you think this is a good idea.

[hadley]

Closing because we're moving towards an approach using .name_spec arguments.