Is there a need to prep a recipe when using vfold_cv, workflow, fit_resample and last_fit?

[kkw]

library(tidyverse)
library(tidymodels)
library(discrim)
library(themis)

df <- read_csv('df.csv')

split <- initial_split(data = df, strata = Pass)
train <- training(split)
test <- testing(split)

df_recipe <- recipes::recipe(Pass ~., data = train) %>%
  step_upsample(Pass) %>%
  recipes::step_normalize(all_predictors()) %>%
  recipes::step_nzv(all_predictors())

set.seed(1)
cv <- vfold_cv(data = train, v = 5, repeats = 5, strata = Pass)

knnmodel <- nearest_neighbor() %>%
  set_engine("kknn") %>%
  set_mode("classification")

knnwf <- workflow() %>%
  add_recipe(df_recipe) %>%
  add_model(knnmodel) 


set.seed(1)
knnres <- knnwf %>%
  tune::fit_resamples(
    resamples = cv,
    metrics = metric_set(roc_auc, sens, spec, accuracy),
    control = control_resamples(save_pred = TRUE, verbose = TRUE)
  )

knnres %>%
  collect_predictions() %>%

final_wf <- workflow() %>%
  add_recipe(df_recipe) %>%
  add_model(knnmodel)


finalres <- final_wf %>%
  last_fit(split)
            

Questions:

1. Do I need to set a separate variable such as

prep_df_recipe <- prep(df_recipe)

In order to use it in workflow?

2. Do I need to pass a preprocessed training data to vfold_cv?

3. Do I need to prep my test dataset as well before last fit?

Thank you!

[Max]

For 1 and 2, you don't need to do anything to the recipe to use fit_resamples(). Give it the untrained recipe.

For number 3, also no. Using the tune and workflows packages, it takes care of all of the details for you.