How does eval and parse get evaluated inside across in dpyr?-CodePudding

I already posted this as an issue in dplyr's repo on github, and they said that across is not a good fit for this type of problem, but I want to post here to see if anyone can give me insight into why this doesn't actually work.

I'm trying to write a function that takes a data frame (y) and uses it to update the column classes of another data frame (x) by matching column names. I had this code lying around and when I wrote it about a year ago I could've swore it worked, but now it doesn't seem to.

library(tidyverse)


(x = tibble(data.frame(a = as.character(c(1,2,3,4)),
                       b = as.character(c(1,2,3,4)),
                       c = as.character(c(1,2,3,4)),
                       d = as.character(c('a', 'b', 'c', 'd')))))
#> # A tibble: 4 × 4
#>   a     b     c     d    
#>   <chr> <chr> <chr> <chr>
#> 1 1     1     1     a    
#> 2 2     2     2     b    
#> 3 3     3     3     c    
#> 4 4     4     4     d

(y = tibble(data.frame(a = as.numeric(c(1,1,1,1)),
                       b = as.character(c(1,1,1,1)),
                       c = as.numeric(c(1,1,1,1)),
                       d = as.character(c('a', 'a', 'a', 'a')))))
#> # A tibble: 4 × 4
#>       a b         c d    
#>   <dbl> <chr> <dbl> <chr>
#> 1     1 1         1 a    
#> 2     1 1         1 a    
#> 3     1 1         1 a    
#> 4     1 1         1 a

## this code i have in my function gives an error
result <-
  x %>%
  dplyr::mutate( dplyr::across( .cols = tidyselect::all_of( colnames( y ) ),
                                .fns  = ~eval(parse(text = paste0(
                                  "as.",
                                  class( y[[dplyr::cur_column()]] ),
                                  "(.)"
                               )))))
#> Error in `dplyr::mutate()`:
#> ! Problem while computing `..1 = dplyr::across(...)`.
#> Caused by error in `across()`:
#> ! Problem while computing column `a`.
#> Caused by error:
#> ! 'list' object cannot be coerced to type 'double'

#> Backtrace:
#>      ▆
#>   1. ├─x %>% ...
#>   2. ├─dplyr::mutate(...)
#>   3. ├─dplyr:::mutate.data.frame(...)
#>   4. │ └─dplyr:::mutate_cols(.data, dplyr_quosures(...), caller_env = caller_env())
#>   5. │   ├─base::withCallingHandlers(...)
#>   6. │   ├─base::withCallingHandlers(...)
#>   7. │   └─mask$eval_all_mutate(quo)
#>   8. ├─base::eval(...)
#>   9. │ └─base::eval(...)
#>  10. └─base::.handleSimpleError(...)
#>  11.   └─dplyr (local) h(simpleError(msg, call))
#>  12.     └─rlang::abort(msg, call = call("across"), parent = cnd)

## and my desired result is:
result = data.frame(a = as.numeric(1,2,3,4),
                    b = as.character(1,2,3,4),
                    c = as.numeric(1,2,3,4),
                    d = as.character('a', 'b', 'c', 'd'))

CodePudding user response：

Or a similar option with across

library(dplyr)
x %>% 
  mutate(across(all_of(names(y)), ~ `class<-`(.x, class(y[[cur_column()]]))))

CodePudding user response：

This seems like a crazy way to achieve the end goal here. A short one-liner using map2_df would do the same thing:

map2_df(x, y, ~ `class<-`(.x, class(.y)))
#> # A tibble: 4 x 4
#>       a b         c d    
#>   <dbl> <chr> <dbl> <chr>
#> 1     1 1         1 a    
#> 2     2 2         2 b    
#> 3     3 3         3 c    
#> 4     4 4         4 d

As for why your code doesn't work, you are right in the sense that this is due to the way that eval works inside a lambda function (it is taking the . as referring to the data frame that was passed into the function call, not as a placeholder to be used inside the lambda function). This is why it is warning you about a list.

If you change the lambda function to a standard function it will work as expected.

x %>%
  dplyr::mutate( dplyr::across( .cols = tidyselect::all_of( colnames( y ) ),
                                .fns  = function(x) eval(parse(text = paste0(
                                  "as.", class(y[[dplyr::cur_column()]]),
                                  "(x)")))
  ))

CodePudding user response：

It is a matter of which environment eval works with. If you specify the current environment then the code in the question works.

  library(dplyr)
  x %>%
    mutate(across(all_of(colnames(y)),
      ~eval(parse(text = paste0("as.", class(y[[cur_column()]]),"(.)"))), 
            environment()))

giving

# A tibble: 4 × 4
      a b         c d    
  <dbl> <chr> <dbl> <chr>
1     1 1         1 a    
2     2 2         2 b    
3     3 3         3 c    
4     4 4         4 d

The code could be simplified by using the as function but you may already know that and are really just looking for info on eval/parse.

x %>%
  mutate(across(all_of(colnames(y)), ~ as(., class(y[[cur_column()]]))))

Note

The inputs simplified:

library(dplyr)
x = tibble(a = 1:4, b = 1:4, c = 1:4, d = letters[1:4])
x[] <- lapply(x, as.character)
y <- tibble(a = rep(1, 4), b = "1", c = 1, d = "a")