purpose of recipes

[jarvmiller]

This looks great. should lead to many good conversations. One thing I'm a bit confused about are two of your "things to note" bullet points when discussing model fitting and prediction, mostly because I have not heard of the recipes package.

• Observation weights and offsets. My thought is that recipes should handle this.
• A subset of the data to work on. Again, recipes should handle this to separate modelling (should be model) fitting and data preprocessing

Just to make sure I understand, do you mean the user does something like so?

my_recipe <- function(df){
  processed <- df %>%
  step_center(all_predictors()) %>%
  step_scale(all_predictors())
  # ... insert 3-5 more steps
  return(processed)
}

knn_family <- new_knn() %>% 
  add_design(my_recipe, data) %>% 
  add_hp_space(hyperparams) %>% 
  add_hp_search(hyperband) %>% 
  add_resampling(resamp_spec) %>% 
  fit()