map_dfr() column binds vectors

[garrettgman]

map_dfr() does the same thing as map_dfc() when given a list of vectors: it column binds the vectors into a data frame.

list_of_vecs <- list(a = c(1,1,1, 1), b = c(2, 2, 2, 2), c = c(3, 3, 3, 3))
list_of_vecs %>% map_dfr(~.x)
## # A tibble: 4 x 3
##       a     b     c
##   <dbl> <dbl> <dbl>
## 1     1     2     3
## 2     1     2     3
## 3     1     2     3
## 4     1     2     3

I'd expect map_dfr() to row bind the vectors into a data frame with one long column, or to throw an error.

[jennybc]

Past thread re: difficulty of row binding: #179

[lionel-]

I'd expect map_dfr() to row bind the vectors into a data frame with one long column, or to throw an error.

Perhaps taking each vector as a row would be more natural? They'd need consistent internal names.

Right now they are taken as columns because bind_rows() and bind_cols() have somewhat sloppy semantics and interpret lists as data frames. This will need to wait on the new vctrs-based tools.

[jennybc]

tibble is also in a holding pattern on a related matter. I might add a discussion of this to my existing slot in Monday's group meeting.

[CottonRockwood]

My expectation would be that map_dfr would r_bind the results as rows if they are vectors while map_dfc would combine the the vectors as columns.

[hadley]

Should switch to vec_rbind() and vec_cbind() and see what breaks.

[leungi]

Faced similar issues when trying to get a tidy tibble from a vector of file paths off fs::dir_ls().

Ended up with purrr::map() + tibble::enframe() + tidyr::unnest() combo.

> library(dplyr)
> 
> file_ls <- fs::dir_ls("./raw_data/Guidelines/",
 regexp = "*.md"
 )
> 
> file_ls %>%
 .[1:2] %>%
 purrr::map_dfr(readr::read_file) %>%
 bind_rows(.id = "doc_id")
# A tibble: 1 x 3
  doc_id `./raw_data/Guidelines/ `./raw_data/Guidelines/
  <chr>  <chr>                                   <chr>                                 
  1      **Coiled~                **Operation~
> 
> file_ls %>%
 .[1:2] %>%
 purrr::map(readr::read_file) %>% 
 tibble::enframe() %>%
 tidyr::unnest()
# A tibble: 2 x 2
  name                                       value                                     
  <chr>                                      <chr>                                     
  ./raw_data/Guidelines/                     **Coiled~
  ./raw_data/Guidelines/                     **Operation~

[iagogv3]

I have a problem I believe it is related to this issue. Let me know if it is. I start with:

mtcars %>%
  split(.$cyl) %>%
  map(~ lm(mpg ~ wt, data = .x)) %>%
  map(summary) %>%
  map_dbl("r.squared")

Let me test a variation. I am using CRAN version of dplyr, so I use yet group_nest instead of nest_by, so later it will be probably easier to do than this:

mtcars %>%
  group_nest(cyl) %>%
  mutate(model = purrr::map(.data$data, ~lm(mpg ~ wt, data = .x)),
         smodl = purrr::map(.data$model, summary),
         ramod = purrr::map_dbl(.data$smodl, "r.squared"),
         aramod = purrr::map_dbl(.data$smodl, "adj.r.squared"))

Then, my goal is: why have one to use 2 times map_dbl inside mutate when it should be possible to use once map_dfc or map_dfr inside bind_cols?

What follows is my attempt to get it. First, I define 3 functions and write in another way the last previous code:

rlm.sq <- function(slmod){
  slmod$r.squared
}
arlm.sq <- function(slmod){
  slmod$adj.r.squared
}
frlm.sq <- function(slmod){
  data.frame(r = slmod$r.squared, a =  slmod$adj.r.squared)
}
mtcars %>%
  group_nest(cyl) %>%
  mutate(model = purrr::map(.data$data, ~lm(mpg ~ wt, data = .x)),
         smodl = purrr::map(.data$model, summary),
         ramod = purrr::map_dbl(.data$smodl, ~rlm.sq(.x)),
         aramod = purrr::map_dbl(.data$smodl, ~arlm.sq(.x)))
# next, a test
mtcars %>%
  group_nest(cyl) %>%
  mutate(model = purrr::map(.data$data, ~lm(mpg ~ wt, data = .x)),
              smodl = purrr::map(.data$model, summary)) %>%
              slice(1) %>%
              pull(smodl) %>%
             extract2(1) %>%
             frlm.sq()

It works, but when I try

mtcars %>%
  group_nest(cyl) %>%
  mutate(model = purrr::map(.data$data, ~lm(mpg ~ wt, data = .x)),
         smodl = purrr::map(.data$model, summary)) %>%
         bind_cols(purrr::map_dfr(.data$smodl, ~frlm.sq(.x)))

I do not get anything. Maybe I should use map_dfc instead of map_dfr?:

mtcars %>%
  group_nest(cyl) %>%
  mutate(model = purrr::map(.data$data, ~lm(mpg ~ wt, data = .x)),
         smodl = purrr::map(.data$model, summary)) %>%
         bind_cols(purrr::map_dfc(.data$smodl, ~frlm.sq(.x)))

Same result.

Actually, in my real code I use neither lm function nor frlm.sq, but, instead of .data$smodl I have a data frame which I summarise to 3 scalar variables besides the grouping variables.

Thank you!

[hadley]

Summary of weirdness:

library(purrr)
c(a = 1, b = 2) %>% map_dfr(~ .x)
#> # A tibble: 1 × 2
#>       a     b
#>   <dbl> <dbl>
#> 1     1     2
c(a = 1, b = 2) %>% map_dfc(~ .x)
#> # A tibble: 1 × 2
#>       a     b
#>   <dbl> <dbl>
#> 1     1     2

1:2 %>% map_dfr(~ .x)
#> Error in `dplyr::bind_rows()`:
#> ! Argument 1 must be a data frame or a named atomic vector.

#> Backtrace:
#>     ▆
#>  1. ├─1:2 %>% map_dfr(~.x)
#>  2. └─purrr::map_dfr(., ~.x)
#>  3.   └─dplyr::bind_rows(res, .id = .id)
#>  4.     └─rlang::abort(glue("Argument {i} must be a data frame or a named atomic vector."))
1:2 %>% map_dfc(~ .x)
#> New names:
#> • `` -> `...1`
#> • `` -> `...2`
#> # A tibble: 1 × 2
#>    ...1  ...2
#>   <int> <int>
#> 1     1     2

1:2 %>% map_dfc(~ setNames(list(.x), letters[.x]))
#> # A tibble: 1 × 2
#>       a     b
#>   <int> <int>
#> 1     1     2
1:2 %>% map_dfr(~ setNames(list(.x), letters[.x]))
#> # A tibble: 2 × 2
#>       a     b
#>   <int> <int>
#> 1     1    NA
#> 2    NA     2

^{Created on 2022-09-02 with reprex v2.0.2}