model-tuninglisted

# Model Tuning Patterns ## Overview Comprehensive hyperparameter optimization using the tune package. Covers grid search, iterative search, racing methods, and Bayesian optimization. ## Tunable Parameters ### Marking Parameters for Tuning ```r library(tidymodels) # Use tune() placeholder in model specification rf_spec <- rand_forest( mtry = tune(), trees = 1000, min_n = tune() ) |> set_engine("ranger") |> set_mode("classification") # Tune recipe steps rec <- recipe(outcome ~ ., data = train) |> step_pca(all_numeric_predictors(), num_comp = tune()) |> step_normalize(all_numeric_predictors()) ``` ### Parameter Objects (dials) ```r library(dials) # View parameter information mtry() min_n() trees() learn_rate() penalty() # Customize parameter ranges mtry(range = c(2, 20)) min_n(range = c(5, 50)) learn_rate(range = c(-3, -1), trans = log10_trans()) # Update based on data mtry_final <- finalize(mtry(), train_data) ``` ## Grid Search ### Regular Grid ```r # Evenly spaced grid regular_grid <- grid_regular( mtry(range = c(2, 10)), min_n(range = c(2, 20)), levels = 5 # 5 levels per parameter = 25 combinations ) # Different levels per parameter regular_grid <- grid_regular( mtry(range = c(2, 10)), min_n(range = c(2, 20)), levels = c(mtry = 5, min_n = 3) ) ``` ### Random Grid ```r # Random sampling of parameter space random_grid <- grid_random( mtry(range = c(2, 10)), min_n(range = c(2, 20)), size = 50 ) ``` ### Space-Filling Designs `