fit_split

library(modeldata)
data("stackoverflow")
library(tidyverse)
library(tidymodels)

set.seed(100) # Important!

# make smaller to save time
  so_split <- initial_split(sample_n(stackoverflow, size = 300), 
                            strata = Remote)
  so_train <- training(so_split)
  so_test  <- testing(so_split)

# again, simpler so runs faster
  so_folds <- vfold_cv(so_train, v = 2, strata = Remote)

# recipe
  tune_rec <- recipe(Remote ~ ., 
                     data = so_train) %>% 
    step_dummy(all_nominal(), -all_outcomes()) %>% 
    step_lincomb(all_predictors()) %>% 
    step_downsample(Remote, under_ratio = tune())

# model
  tune_spec <-
    rand_forest(mtry = tune(),
                min_n = tune()) %>% 
    set_engine("ranger") %>% 
    set_mode("classification")

# workflow
  tuneboth_wf <-
  workflow() %>% 
    add_recipe(tune_rec) %>% 
    add_model(tune_spec)

# tuning parameters
  tuneboth_param <- parameters(tuneboth_wf)

  tuneboth_param <- 
  tuneboth_param %>% 
  # Pick an upper bound for mtry: 
    update(mtry = mtry(c(1, 20)))

# fit
  results <- 
  tuneboth_wf %>% 
    tune_grid(resamples = so_folds, 
              param_info = tuneboth_param)

# get best tuning parameters
  best <- 
  results %>% 
    select_best(metric = "roc_auc")

# define final workflow
  wf_final <- 
    tuneboth_wf %>%
    finalize_workflow(best)

# re-run with best
  fit_split <- function(formula, model, split, ...) {
    wf <- workflows::add_model(
      workflows::add_formula(workflows::workflow(), 
                             formula, 
                             blueprint = 
                               hardhat::default_formula_blueprint(
                                 indicators = FALSE, 
                                 allow_novel_levels = TRUE)), 
      model)
    tune::last_fit(wf, split, ...)
  }
  
results_best_test <-
  wf_final %>%
  fit_split(split = so_split,
            metrics = metric_set(roc_auc, sens, spec))