使用嵌套数据框访问purrr :: map（）中的分组变量

Question

I'm using tidyr::nest() in combination with purrr::map() (-family) to group a data.frame into groups and then do some fancy stuff with each subset. Consider following example, and please ignore the fact that I don't need nest() and map() to do this (this is an oversimplified example):

library(dplyr)
library(purrr)
library(tidyr)

mtcars %>% 
  group_by(cyl) %>%
  nest() %>%
  mutate(
    wt_mean = map_dbl(data,~mean(.x$wt))
  )

# A tibble: 8 x 4
    cyl  gear data               cly2
  <dbl> <dbl> <list>            <dbl>
1     6     4 <tibble [4 x 9]>      6
2     4     4 <tibble [8 x 9]>      4
3     6     3 <tibble [2 x 9]>      6
4     8     3 <tibble [12 x 9]>     8
5     4     3 <tibble [1 x 9]>      4
6     4     5 <tibble [2 x 9]>      4
7     8     5 <tibble [2 x 9]>      8
8     6     5 <tibble [1 x 9]>      6

Usually when I do this type of operation, I need access to the grouping variable ( cyl in this case) within map() . But these grouping variables appear as vectors with length corresponding to the number of rows in the nested dataframe, and therefore don't lend themselves easily.

Is there a way I could run the following operation? I would want the mean of wt to be divided by the number of cylinders ( cyl ) per group (ie row).

mtcars %>% 
  group_by(cyl,gear) %>%
  nest() %>%
  mutate(
    wt_mean = map_dbl(data,~mean(.x$wt)/cyl)
  )


Error in mutate_impl(.data, dots) : 
  Evaluation error: Result 1 is not a length 1 atomic vector.

Answer 1

Take cyl out of the map call:

mtcars %>% 
  group_by(cyl,gear) %>%
  nest() %>%
  mutate(
    wt_mean = map_dbl(data, ~mean(.x$wt)) / cyl
  )

# A tibble: 8 x 4
    cyl  gear data              wt_mean
  <dbl> <dbl> <list>              <dbl>
1     6     4 <tibble [4 x 9]>    0.516
2     4     4 <tibble [8 x 9]>    0.595
3     6     3 <tibble [2 x 9]>    0.556
4     8     3 <tibble [12 x 9]>   0.513
5     4     3 <tibble [1 x 9]>    0.616
6     4     5 <tibble [2 x 9]>    0.457
7     8     5 <tibble [2 x 9]>    0.421
8     6     5 <tibble [1 x 9]>    0.462

map_dbl sees cyl as a length 8 vector because nest removes groups from data.frame . Using cyl in map_* function call (as in OP's example) results in 8 length-8 vectors.

2 other approaches:

Both with same result as above, but keep the grouped variables in the map_* call, per OP's specs:

Re grouping after nest

mtcars %>% 
  group_by(cyl,gear) %>%
  nest() %>%
  group_by(cyl, gear) %>%
  mutate(wt_mean = map_dbl(data,~mean(.x$wt)/cyl))

map2 for iterating over cyl

mtcars %>% 
  group_by(cyl,gear) %>%
  nest() %>%
  mutate(wt_mean = map2_dbl(data, cyl,~mean(.x$wt)/ .y))

Answer 2

In the new release of dplyr 0-8-0 , you can now use group_map , which I find very handy for this use case. This is the example by github user @yutannihilation

library(dplyr, warn.conflicts = FALSE)

mtcars %>% 
  group_by(cyl) %>%
  group_map(function(data, group_info) {
    tibble::tibble(wt_mean = mean(data$wt) / group_info$cyl)
  })