Nrthugu

I'm trying to tighten up a %>% piped workflow where I need to apply the same function to several columns but with one argument changed each time. I feel like purrr's map or invoke functions should help, but I can't wrap my head around it.

My data frame has columns for life expectancy, poverty rate, and median household income. I can pass all these column names to vars in mutate_at, use round as the function to apply to each, and optionally supply a digits argument. But I can't figure out a way, if one exists, to pass different values for digits associated with each column. I'd like life expectancy rounded to 1 digit, poverty rounded to 2, and income rounded to 0.

I can call mutate on each column, but given that I might have more columns all receiving the same function with only an additional argument changed, I'd like something more concise.

library(tidyverse)



df <- tibble::tribble(

        ~name, ~life_expectancy,          ~poverty, ~household_income,

  "New Haven", 78.0580437642378, 0.264221051111753,  42588.7592521085

  )

In my imagination, I could do something like this:

df %>%

  mutate_at(vars(life_expectancy, poverty, household_income), 

            round, digits = c(1, 2, 0))

But get the error

Error in mutate_impl(.data, dots) :
Column life_expectancy must be length 1 (the number of rows), not 3

Using mutate_at instead of mutate just to have the same syntax as in my ideal case:

df %>%

  mutate_at(vars(life_expectancy), round, digits = 1) %>%

  mutate_at(vars(poverty), round, digits = 2) %>%

  mutate_at(vars(household_income), round, digits = 0)

#> # A tibble: 1 x 4

#>   name      life_expectancy poverty household_income

#>   <chr>               <dbl>   <dbl>            <dbl>

#> 1 New Haven            78.1    0.26            42589

Mapping over the digits uses each of the digits options for each column, not by position, giving me 3 rows each rounded to a different number of digits.

df %>%

  mutate_at(vars(life_expectancy, poverty, household_income), 

            function(x) map(x, round, digits = c(1, 2, 0))) %>%

  unnest()

#> # A tibble: 3 x 4

#>   name      life_expectancy poverty household_income

#>   <chr>               <dbl>   <dbl>            <dbl>

#> 1 New Haven            78.1    0.3            42589.

#> 2 New Haven            78.1    0.26           42589.

#> 3 New Haven            78      0              42589

^{Created on 2018-11-13 by the reprex package (v0.2.1)}

2 solutions

mutate with !!!

invoke was a good idea but you need it less now that most tidyverse functions support the !!! operator, here's what you can do :

digits <- c(life_expectancy = 1, poverty = 2, household_income = 0)  

df %>% mutate(!!!imap(digits, ~round(..3[[.y]], .x),.))

# # A tibble: 1 x 4

#          name life_expectancy poverty household_income

#         <chr>           <dbl>   <dbl>            <dbl>

#   1 New Haven            78.1    0.26            42589

..3 is the initial data frame, passed to the function as a third argument, through the dot at the end of the call.

Written more explicitly :

df %>% mutate(!!!imap(

  digits, 

  function(digit, name, data) round(data[[name]], digit),

  data = .))

If you need to start from your old interface (though the one I propose will be more flexible), first do:

digits <- setNames(c(1, 2, 0), c("life_expectancy", "poverty", "household_income"))

mutate_at and <<-

Here we bend a bit the good practice of avoiding <<- whenever possible, but readability matters and this one is really easy to read.

digits <- c(1, 2, 0)

i <- 0

df %>%

  mutate_at(vars(life_expectancy, poverty, household_income), ~round(., digits[i<<- i+1]))

# A tibble: 1 x 4

#     name      life_expectancy poverty household_income

#     <chr>               <dbl>   <dbl>            <dbl>

#   1 New Haven            78.1    0.26            42589

(or just df %>% mutate_at(names(digits), ~round(., digits[i<<- i+1])) if you use a named vector as in my first solution)

Fun with tidyeval:

prepared_pairs <- 

  map2(

    set_names(syms(list("life_expectancy", "poverty", "household_income"))),

    c(1, 2, 0), 

    ~expr(round(!!.x, digits = !!.y))

  )



mutate(df, !!! prepared_pairs)



# # A tibble: 1 x 4

#   name      life_expectancy poverty household_income

#   <chr>               <dbl>   <dbl>            <dbl>

# 1 New Haven            78.1    0.26            42589

Here's a map2 solution along the lines of Henrik's comment. You can then wrap this inside a custom function. I provided an rough first attempt but I have done minimal tests, so it probably breaks under all sorts of situations if evaluation is strange. It also doesn't use tidyselect for .at, but neither does modify_at...

library(tidyverse)



df <- tibble::tribble(

  ~name, ~life_expectancy,          ~poverty, ~household_income,

  "New Haven", 78.0580437642378, 0.264221051111753,  42588.7592521085,

  "New York", 12.349685329, 0.324067934, 32156.230974623

)



rounded <- df %>%

  select(life_expectancy, poverty, household_income) %>%

  map2_dfc(

    .y = c(1, 2, 0),

    .f = ~ round(.x, digits = .y)

  )

df %>%

  select(-life_expectancy, -poverty, -household_income) %>%

  bind_cols(rounded)

#> # A tibble: 2 x 4

#>   name      life_expectancy poverty household_income

#>   <chr>               <dbl>   <dbl>            <dbl>

#> 1 New Haven            78.1    0.26            42589

#> 2 New York             12.3    0.32            32156





modify2_at <- function(.x, .y, .at, .f) {

  modified <- .x[.at] %>%

    map2(.y, .f)

  .x[.at] <- modified

  return(.x)

}



df %>%

  modify2_at(

    .y = c(1, 2, 0),

    .at = c("life_expectancy", "poverty", "household_income"),

    .f = ~ round(.x, digits = .y)

  )

#> # A tibble: 2 x 4

#>   name      life_expectancy poverty household_income

#>   <chr>               <dbl>   <dbl>            <dbl>

#> 1 New Haven            78.1    0.26            42589

#> 2 New York             12.3    0.32            32156

^{Created on 2018-11-13 by the reprex package (v0.2.1)}