Passing an argument a column name to mean function inside a custom function

Question

I want to perform a mean encoding to a high-cardinality variable using 5-fold cross validation. My code is:

df <- data.frame(sample(c(1,2,3,4,5), 1000, replace=T), sample(c(1,0), 1000, replace=T))
colnames(df) <- c("var", "target")

encode <- function(df, target_var, column_var){

set.seed(520)
df$group <- as.factor(sample(c(1,2,3,4,5), nrow(df), replace=T, prob=c(0.2,0.2,0.2,0.2,0.2)))

var.enc <- df %>% 
               select_("group", column_var, target_var) %>% 
               group_by_("group", column_var) %>% 
               mutate(var_encoded = mean(target_var)) %>% 
               ungroup() %>% 
               select_(column_var, "var_encoded") %>% 
               distinct() %>% 
               group_by_(column_var) %>% 
               mutate(var.enc = mean(var_encoded)) %>% 
               distinct()

return(var.enc)
}

encoding <- encode(df = df, column_var = "var", target_var = "target")

When I run the code above I got an warning:

In mean.default(target_var) : argument is not numeric or logical: returning NA

So, how can I pass the argument correctly to mean inside my function? I have tried to use as.name() but does not work either. Also, I have used mean(df[[target_var]]), but this way the group_by is not working and I am getting the global mean as a result.

EDIT: I have added a reproducible example.

Answer 1

As the input is a character string, convert it to symbol (sym) and then do the evaluate (!!)

encode <- function(df, target_var, column_var){

  set.seed(520)
  df$group <- as.factor(sample(c(1,2,3,4,5), nrow(df),
         replace=T, prob=c(0.2,0.2,0.2,0.2,0.2)))

  column_var <- rlang::sym(column_var)
  target_var <- rlang::sym(target_var)
  df %>% 
       select(group, !! column_var, !!target_var) %>%  
        group_by(group, !! column_var) %>% 
        mutate(var_encoded = mean(!!target_var)) %>% 
        ungroup() %>% 
        select(!! column_var, var_encoded) %>% 
        distinct() %>% 
        group_by(!! column_var) %>% 
        mutate(var.enc = mean(var_encoded)) %>% 
       distinct()


}

-checking

encoding <- encode(df = df, target_var = "target", column_var = "var")
encoding
# A tibble: 25 x 3
# Groups:   var [5]
#     var var_encoded var.enc
#   <dbl>       <dbl>   <dbl>
# 1     5       0.462   0.497
# 2     5       0.553   0.497
# 3     4       0.585   0.493
# 4     2       0.543   0.536
# 5     3       0.364   0.453
# 6     4       0.46    0.493
# 7     1       0.465   0.476
# 8     3       0.474   0.453
# 9     5       0.529   0.497
#10     1       0.417   0.476
# ... with 15 more rows