如何用Tidymodels和vip绘制随机森林的变量重要性图

我使用的数据集为蘑菇数据集。

• 已通过Tidymodels完成随机森林的配方、模型及工作流定义：

df_recipe_mixt <- df_train |>  recipe(class ~ cap_diameter + cap_color + does_bruise_or_bleed + gill_color + stem_height + stem_width + stem_color + has_ring + habitat + season, data = df_train) |>
  step_scale(all_numeric()) |> 
  step_center(all_numeric()) |> 
  step_dummy(all_nominal(), -all_outcomes()) |> 
  prep()

rf_mod <- rand_forest() |> 
  set_engine("ranger") |> 
  set_mode("classification") |> 
  set_args(mtry = tune(), trees = tune())

rf_wf <- workflow() |>  
  add_model(rf_mod) |> 
  add_recipe(df_recipe_mixt) 

• 已完成模型调优并应用至测试集：

n_cores <- parallel::detectCores(logical = TRUE)
registerDoParallel(cores = n_cores - 1)

rf_params <- extract_parameter_set_dials(rf_wf) |>  
  update(mtry = mtry(c(1,5)), trees = trees(c(50,500)))

rf_grid <- grid_regular(rf_params, levels = c(mtry = 5, trees = 3))

tic("random forest model tuning ")

tune_res_rf <- tune_grid(rf_wf,
  resamples = df_folds,
  grid = rf_grid,
  metrics = metric_set(accuracy)
)

toc()

stopImplicitCluster()

autoplot(tune_res_rf) + dark_mode(theme_minimal())

rf_best <- tune_res_rf |> select_best(metric = "accuracy")

rf_best$trees;rf_best$mtry

rf_final_wf <- rf_wf |>
  finalize_workflow(rf_best)

rf_res <- last_fit(rf_final_wf, split = df_split) |> collect_predictions()

模型运行正常，已获取所有指标、混淆矩阵及ROC曲线，但不清楚如何使用vip包绘制变量重要性图。

内容的提问来源于stack exchange，提问作者Smorg