The goal is to be able to add a series of new steps in sequence to the end of a recipe where the new steps are passed as arguments to a function.
I can do what I want with a for
loop:
library(tidyverse)
suppressPackageStartupMessages(library(recipes))
suppressPackageStartupMessages(library(rlang))
add_steps_to_recipe <- function(rec, new_steps) {
rec_new <- rec
for (i in seq_along(new_steps)) {
rec_new <- eval_tidy(expr(rec_new %>% !!new_steps[[i]]))
}
return(rec_new)
}
.pred_date <- as.Date("2015-01-15")
mtcars2 <- mtcars %>% mutate(hp_date = as.Date("2015-01-01"))
mtcars2$hp_date[1:2] <- as.Date("2015-02-01")
rec1 <- recipe(mtcars2, mpg ~ hp + hp_date)
new_steps <- exprs(
step_mutate(hp = ifelse(hp_date < .pred_date, hp, as.numeric(NA))),
step_meanimpute(hp)
)
rec2 <- add_steps_to_recipe(rec1, new_steps)
juice(prep(rec2))
I can't seem to get it to work without using a for
loop:
new_steps <- expr(
step_mutate(hp = ifelse(hp_date < .pred_date, hp, as.numeric(NA))) %>% step_meanimpute(hp)
)
add_steps_to_recipe <- function(rec, new_steps) {
eval_tidy(expr(rec %>% !!new_steps))
}
rec2 <- add_steps_to_recipe(rec1, new_steps)
#> Error in `%>%`(., step_mutate(hp = ifelse(hp_date < .pred_date, hp, as.numeric(NA))), : unused argument (step_meanimpute(hp))
juice(prep(rec2))
#> Error in prep(rec2): object 'rec2' not found
Created on 2020-02-24 by the reprex package (v0.3.0)
The problem is caused by the pipe in new_steps
. If I just add one step, it works fine. I noticed also that the unquoted expression has an additional set of parentheses. Is there a way to remove an extra set of parentheses after unquoting?