Trying to use tidymodels for a catboost model: Receiving error related to labels

Question

Here is the model:

cb_spec <- boost_tree(
  mode = "classification",
  trees = 1000,
  tree_depth = tune(),
  min_n = tune(),
  mtry = tune(),
  learn_rate = tune()
) %>%
  set_engine("catboost", loss_function = "Logloss", task_type = "GPU")

Here is the recipe:

cb_rec <- recipe(covid_vaccination ~ ., data = cb_train) %>% 
  step_unknown(all_nominal_predictors()) %>%
  #step_dummy(all_nominal_predictors(), one_hot = TRUE) %>%
  step_impute_median(all_numeric_predictors()) %>% 
  step_nzv(all_predictors())

I combine them:

cb_wf <- workflow() %>% 
  add_model(cb_spec) %>% 
  add_recipe(cb_rec)

Then I try to tune to find optimal hyperparameters:

cb_tune <- tune_grid(
  object = cb_wf,
  resamples = cb_folds,
  grid = cb_grid,
  metrics = metric_set(roc_auc),
  control = control_grid(verbose = TRUE)
)

Here is the error I get:

Error in catboost.from_matrix(as.matrix(float_and_cat_features_data), : Unsupported label type, expecting double or integer. I have already confirmed that categorical variables are changed to factors. There are absolutely no character type vectors in my dataset.

Answer 1

This was confirmed to be a weird error when using tidymodels for catboost. Check their github issues for more info and a current workaround.