making loop-friendly formula interface in tidyeval functions

user_7973626 0
Viewed 69

I am writing a simple function using tidyeval where I need to pass the arguments to the formula interface. Although I have managed to build a working version of the function, it doesn't seem to work with for loops.

function

foo <- function(data, x, y) {
  BayesFactor::ttestBF(
    paired = FALSE,
    data = data,
    formula = rlang::new_formula(rlang::enexpr(y), rlang::enexpr(x))
  )
}

foo(mtcars, am, wt)
#> Bayes factor analysis
#> --------------
#> [1] Alt., r=0.707 : 1383.367 ±0%
#> 
#> Against denominator:
#>   Null, mu1-mu2 = 0 
#> ---
#> Bayes factor type: BFindepSample, JZS

working with loops

I also tried here !!col.name[i]

df <- dplyr::select(mtcars, am, wt, mpg)
col.name <- colnames(df)

for (i in 2:length(col.name)) {
  foo(
    data = mtcars,
    x = am,
    y = col.name[i]
  )
}
#> Error in `[.data.frame`(data, , dv): undefined columns selected
tidyeval r
2 Answers

If you want to make a data-masking function work with loops over columns, you have to do metaprogramming at some point.

Really there are two options:

  • Either make your function take strings with standard evaluation. Then transform that string to a symbol internally. The metaprogramming is internal.

  • Or make it take expressions with non-standard evaluation. Then your callers have to transform strings to symbols and unquote them. The metaprogramIng is external.

There is no way around that, unless you're going to create a non standard interface that works inconsistently and unpredictably by trying to be too magical.

user_17251770

If you can pass string values as variables we can use reformulate to construct the formula.

foo <- function(data, x, y) {
  BayesFactor::ttestBF(
    paired = FALSE,
    data = data,
    formula = reformulate(x, y)
  )
}

foo(mtcars, "am", "wt")

#Bayes factor analysis
#--------------
#[1] Alt., r=0.707 : 1383.367294 ±0%

#Against denominator:
#  Null, mu1-mu2 = 0 
#---
#Bayes factor type: BFindepSample, JZS

To pass it in a loop/lapply :

col.name <- c('wt', 'mpg')
result <- lapply(col.name, foo, data = mtcars, x = 'am')
result

#[[1]]
#Bayes factor analysis
#--------------
#[1] Alt., r=0.707 : 1383.367294 ±0%

#Against denominator:
#  Null, mu1-mu2 = 0 
#---
#Bayes factor type: BFindepSample, JZS


#[[2]]
#Bayes factor analysis
#--------------
#[1] Alt., r=0.707 : 86.58972736 ±0%

#Against denominator:
#  Null, mu1-mu2 = 0 
#---
#Bayes factor type: BFindepSample, JZS
user_39629140
Related