mutate using fn that returns a df or list

[cawthm]

Suppose we have a function that returns a named df/list with more than one variable output.

Question: How can run that function against a df and create more than one new variable at a time, and what is the most natural way with dplyr/ purrr?

library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(purrr)

toy_fun <- function(chr) {
  data.frame(let_rank = which(letters == chr),
             rando = rnorm(n = length(chr)))
}

# for example, given

df <- data.frame(lets = sample(letters, 10))

# what is the correct dplyr/purrr way to map the function to 
# capture the output of the function?

# this of course fails

df %>% mutate(toy_fun(lets), .id = names(toy_fun(lets)))
#> Error in mutate_impl(.data, dots): Evaluation error: argument "chr" is missing, with no default.

# here's one way which works, but which seems unwieldy and ugly:

bind_cols(df, map_df(df$lets, toy_fun))
#>    lets let_rank       rando
#> 1     d        4 -0.30910301
#> 2     j       10  1.54324912
#> 3     i        9 -0.57664505
#> 4     u       21  1.15671969
#> 5     b        2 -0.03828406
#> 6     f        6 -0.64202232
#> 7     l       12  0.50793796
#> 8     y       25  0.98867755
#> 9     x       24  0.02617367
#> 10    h        8 -0.75882107

# this doesn't work like I thought it would:
map_dfc(df$lets, toy_fun)
#>   let_rank     rando let_rank1   rando1 let_rank2   rando2 let_rank3
#> 1        4 0.8284374        10 1.342487         9 2.982913        21
#>     rando3 let_rank4    rando4 let_rank5    rando5 let_rank6   rando6
#> 1 0.568951         2 0.7849361         6 -0.180958        12 2.808249
#>   let_rank7    rando7 let_rank8       rando8 let_rank9     rando9
#> 1        25 0.4196204        24 -0.008707815         8 -0.5353852

^{Created on 2018-11-22 by the reprex package (v0.2.1)}

[mishabalyasin]

I would say, that common way to do it is map->unnest combo. It is how you would use it in, for example, broom.

suppressPackageStartupMessages(library(tidyverse))

toy_fun <- function(chr) {
  data.frame(let_rank = which(letters == chr),
             rando = rnorm(n = length(chr)))
}
df <- data.frame(lets = sample(letters, 10))
df %>% 
  mutate(res = purrr::map(lets, toy_fun)) %>% 
  tidyr::unnest(res)
#>    lets let_rank      rando
#> 1     l       12  0.5670311
#> 2     s       19  0.6504350
#> 3     k       11 -1.0761233
#> 4     n       14 -0.8507612
#> 5     x       24  0.2007877
#> 6     y       25 -0.3233344
#> 7     t       20  0.5016116
#> 8     h        8  0.8095949
#> 9     q       17 -1.4610645
#> 10    p       16  0.7157577

^{Created on 2018-11-22 by the reprex package (v0.2.1)}

[jcblum]

These links from a couple of related discussions might be of interest (expand the box to see the links in the first one):

Split-apply-combine approaches

I think you could be interested in these resource as well You can do with dplyr and purrr what you could do with group_by() and do.

Why doesn't group_by %>% purr::f() behave the same way as summarise()?

The new DataCamp course, Machine Learning in the Tidyverse, is both excellent and makes extensive use of list-columns! Even if you don't have a DataCamp membership, the first chapter is available for free and could be a good place to start. From the description of Chapter 1: This chapter will introduce you to the backbone of machine learning in the tidyverse, the List Column Workflow (LCW). The LCW will empower you to work with many models in one dataframe.