Skip to contents

Get default hyperparameter grids for supported survival models

Source: R/machine_learning.R
get_default_hyperparams.Rd

This function generates default hyperparameter grids for various survival models compatible with the tidymodels framework. It uses the default parameter ranges from the dials package and produces a sequence of evenly spaced values within those ranges for each tunable hyperparameter.

Usage

get_default_hyperparams(model_name, train_x = NULL, levels = 5)

Arguments

model_name

Character string specifying the model name. Supported options include:

  • "cox_ph_survival" – Classic Cox proportional hazards model

  • "proportional_hazards_glmnet" – Penalized Cox (LASSO / Elastic Net)

  • "survreg_flexsurv" – Parametric accelerated failure time (AFT)

  • "decision_tree_partykit" – Single survival tree

  • "bag_tree_rpart" – Bagged CART survival trees

  • "rand_forest_partykit" – Random survival forest (ctree-based)

  • "rand_forest_aorsf" – Oblique random survival forest

  • "boost_tree_mboost" – Gradient boosting for survival

train_x

Optional data frame or matrix of training predictors. This is required for parameters that depend on the number of features, such as mtry, which determines the number of variables randomly sampled at each split in tree-based models.

levels

Integer specifying how many values to generate per hyperparameter. Defaults to 5. Must be at least 2.

Value

A named list of hyperparameter grids. Each element is a numeric vector of sampled values for that parameter. Returns NULL for models without tunable hyperparameters.

Details

The function supports models such as Cox proportional hazards (regular and penalized), parametric survival regression, decision trees, bagging, random forests, and gradient boosting. For models without tunable parameters (e.g., classic Cox or AFT models), the function returns NULL.

The helper function vs() internally calls dials::value_seq() to generate evenly spaced sequences of parameter values across their default ranges. For data-dependent parameters (like mtry), dials::finalize() is used to compute appropriate limits based on train_x.

Examples

# Example 1: Random forest with feature-dependent hyperparameter
get_default_hyperparams("rand_forest_partykit", train_x = iris[, 1:4])
#> $trees
#> [1]    1  500 1000 1500 2000
#> 
#> $min_n
#> [1]  2 11 21 30 40
#> 
#> $mtry
#> [1] 1 2 3 4
#> 

# Example 2: Penalized Cox model with default parameter ranges
get_default_hyperparams("proportional_hazards_glmnet")
#> $penalty
#> [1] 1.000000e-10 3.162278e-08 1.000000e-05 3.162278e-03 1.000000e+00
#> 
#> $mixture
#> [1] 0.00 0.25 0.50 0.75 1.00
#>