export magrittr::extract()?

[jennybc]

I taught purrr today. We selected multiple elements at the same level of the list hierarchy. Overall reaction was "elegant and simplifying", but mapping [ was described as "ugly and horrifying".

Is there any chance of re-exporting magrittr::extract()?

library(purrr)
library(magrittr)

x <- list(list(one = list(alpha = 1, beta = 2), two = "three", three = 4),
          list(one = list(alpha = 5, beta = 6), two = "seven", three = 8))
#x %>% map_dbl(c("one", "beta"))
#x %>% map_df(`[`, c("two", "three"))
x %>% map_df(extract, c("two", "three"))
#> # A tibble: 2 x 2
#>     two three
#>   <chr> <dbl>
#> 1 three     4
#> 2 seven     8

[lionel-]

mapping [ was described as "ugly and horrifying"

:) That may just be because they are not used to it. Once you get the hang of mapping operators you can do stuff like this: x %>% some(==, "") which is kind of elegant I would say.

extract() is already quite overloaded across packages. Maybe vct_subset() and vct_extract() for [and [[ respectively? I like the vct_ prefix (or maybe vec_) but it requires knowing that both lists and atomic vectors are vectors and it's not how R neophytes usually think about it.

[jennybc]

That may just be because they are not used to it

That's true. My own horror at this has now faded to a distant memory.

I think it catches attention because many people don't predict that using a vector as .f will index at successively deeper levels of the input, e.g. map_dbl(c("one", "beta")) in the example. That's not very well-known. So it's easy to be surprised about which syntax does what. And to be disappointed about which one requires a bit more work.

retrieve()?

Also in a past email conversation, there was talk of a select method for lists. Reminding you of a relevant example you wrote:

library(purrr)
library(dplyr)
x <- rerun(2, head(mtcars,2))
x %>% map_df(select, cyl, am)
#>   cyl am
#> 1   6  1
#> 2   6  1
#> 3   6  1
#> 4   6  1

[hadley]

I'd rather fix this by improving the documentation for map() so that you know you can use a character vector to index deeply into an object.

[jennybc]

I never really thought this would fly but just want to point out this issue was about retrieving "multiple elements at the same level of the list hierarchy". So, more like dplyr::select() but for lists. It was explicitly NOT about using "a character vector to index deeply into an object", which is already well documented.

[hadley]

Oh gotcha. I think an explicit [ is probably still the best way to do that even if it makes your students retch in horror.

[dmi3kno]

The issue with explicit [, as indicated here and here is that it does not have crucial .default/.null handling. I would insist that the issue is still open and there's a need for multi-value pluck for "tidying" lists.

> purrr::map_dfr(x, ~list(two = purrr::pluck(.x, "two", .default = NA), 
                         three = purrr::pluck(.x, "three", .default = NA)))
#> or

> purrr::map_dfr(x, function(i) purrr::map(purrr::set_names(c("two", "three")), 
                                          ~purrr::pluck(i, .x, .default = NA)))

#>  A tibble: 2 x 2
#>  two   three
#>  <chr> <dbl>
#>1 three  4.00
#>2 seven  8.00

Or, what I usually need for creating list-columns:

> purrr::map(x, ~tibble::tibble(two=purrr::pluck(.x, "two", .default = NA), 
                      three=purrr::pluck(.x, "three", .default = NA)))
#> or

> purrr::map(x, function(i) purrr::map_dfr(purrr::set_names(c("two", "three")), 
                                        ~purrr::pluck(i, .x, .default = NA)))

#> [[1]]
#> # A tibble: 1 x 2
#>   two   three
#>   <chr> <dbl>
#> 1 three  4.00
#> 
#> [[2]]
#> # A tibble: 1 x 2
#>   two   three
#>   <chr> <dbl>
#> 1 seven  8.00

[batpigandme]

@dmi3kno since this issue was closed over a year ago, commenting here is pretty low-visibility. If you could please open a new issue that refers back to this with your reprex etc., it will be much easier for the maintainers to find it.

Thanks