Benchmark LiblineaR and glmnet engines for logistic regression in tidymodels

benchmark_logistic_reg.R

library(tidymodels)
library(textrecipes)

data("small_fine_foods")
sparse_bp <- hardhat::default_recipe_blueprint(composition = "dgCMatrix")

text_rec <-
  recipe(score ~ review, data = training_data) %>%
  step_tokenize(review)  %>%
  step_stopwords(review) %>%
  step_tokenfilter(review, max_tokens = 1e3) %>%
  step_tfidf(review)

liblinear_spec <- logistic_reg(penalty = 0.02, mixture = 1) %>%
  set_engine("LiblineaR") %>%
  set_mode("classification")

glmnet_spec <- logistic_reg(penalty = 0.02, mixture = 1) %>%
  set_engine("glmnet") %>%
  set_mode("classification")

liblinear <- 
  workflow() %>%
  add_recipe(text_rec) %>%
  add_model(liblinear_spec)

glmnet_sparse <- 
  workflow() %>%
  add_recipe(text_rec, blueprint = sparse_bp) %>%
  add_model(glmnet_spec)

glmnet_default <- 
  workflow() %>%
  add_recipe(text_rec) %>%
  add_model(glmnet_spec)


set.seed(123)
food_folds <- vfold_cv(training_data, v = 3)

library(bench)

results <- mark(
  iterations = 10, check = FALSE,
  liblinear = fit_resamples(liblinear, food_folds), 
  glmnet_sparse = fit_resamples(glmnet_sparse, food_folds),  
  glmnet_default = fit_resamples(glmnet_default, food_folds), 
)

results

autoplot(results, type = "ridge")