CODEWITHSUNDEEP
rtidymodels
te time
te time

Reputation: 495

Implementing tune_bayes in Tidymodels workflow_sets

I am trying to build a workflow_set using tidymodels to compare models that are tuned using the bayesian approach. I am not sure how to map the parameters that need to be tuned into the workflow_set. I am getting the following error

"Error in dials::grid_latin_hypercube(param, size = n) : \n  These arguments contains unknowns: `mtry`. See the `finalize()` function.\n"

It looks like I need to supply initial values when not using a workflow_set I would do

    xgb_wflow <-
   tree_frogs_wflow %>%
   add_model(xgb_spec)
 
   xgb_params <- extract_parameter_set_dials(xgb_spec)%>%
   finalize(tree_frogs_train)
   xgb_res <- 
   tune_bayes(
              object = xgb_wflow,
              resamples = folds,
              param_info=xgb_params,
              metrics=metric_set(pr_auc,roc_auc,accuracy),
              initial = 10,
              control =  ctrl_bayes)

not sure how to do this using workflow_sets. My attempt is below any help is appreciated.

library(tidymodels)
library(tidyverse)
library(stacks) 

data("tree_frogs")

tree_frogs <- tree_frogs %>%
  select(-c(clutch, latency)) 
set.seed(1)

tree_frogs_split <- initial_split(tree_frogs)
tree_frogs_train <- training(tree_frogs_split)
tree_frogs_test  <- testing(tree_frogs_split)

tree_frogs_rec <- 
  recipe(reflex ~ ., data = tree_frogs_train) %>%
  step_dummy(all_nominal(), -reflex) %>%
  step_zv(all_predictors())


set.seed(1)
folds <- rsample::vfold_cv(tree_frogs_train, v = 5)


## multinomial 


mlt_spec <- 
  multinom_reg(penalty = tune(), mixture = 1) %>% 
  set_engine("glmnet") %>% 
  set_mode("classification")

## Random Forest 

rand_forest_spec <- 
  rand_forest(
    mtry = tune(),
    min_n = tune(),
    trees = 500
  ) %>%
  set_mode("classification") %>%
  set_engine("ranger")


## XGBoost 

xgb_spec <- 
  boost_tree(
    trees = 1000,
    min_n = tune(),
    learn_rate = tune(),
    loss_reduction = tune(),
    sample_size = tune(),
    mtry = tune(),
    tree_depth = tune()
  ) %>%
  set_engine("xgboost") %>%
  set_mode("classification")


all_workflows <- 
  workflow_set(
    preproc = list("basic_rec" = tree_frogs_rec),
    models = list(mlt = mlt_spec, rf = rand_forest_spec, xgb = xgb_spec)
  )

ctrl_bayes = control_bayes(save_pred = TRUE,
                           parallel_over = "everything",
                           save_workflow = TRUE)

bayes_results <-
  all_workflows %>%
  workflow_map(
    seed = 1,
    fn = "tune_bayes",
    resamples = folds,
    initial = 10,
    control = ctrl_bayes
  )

bayes_results %>% 
  rank_results() %>% 
  filter(.metric == "pr_auc") %>% 
  select(model, .config, pr_auc = mean, rank)

Upvotes: 0

Views: 416

Answers (1)

te time
te time

Reputation: 495

I found the answer in the documentation using the options_add function

I added the code below before the workflow_map

mlt_params <- extract_parameter_set_dials(mlt_spec)%>%
  finalize(tree_frogs_train)

xgb_params <- extract_parameter_set_dials(xgb_spec)%>%
  finalize(tree_frogs_train)

rf_params <- extract_parameter_set_dials(rand_forest_spec)%>%
  finalize(tree_frogs_train)

all_workflows = all_workflows %>%
  option_add(param_info = mlt_params,id = "basic_rec_mlt")

all_workflows = all_workflows %>%
  option_add(param_info = rf_params,id = "basic_rec_rf")

all_workflows = all_workflows %>%
  option_add(param_info = xgb_params,id = "basic_rec_xgb")

Upvotes: 2

Related Questions