Train the Best Survival Model Using Optimized Hyperparameters

Fits a survival model on the full training data using the optimal hyperparameters obtained from nested cross-validation. This wrapper ensures consistent retraining for different survival model types (Cox, penalized Cox, AFT, tree-based, or ensemble models), and supports preprocessing pipelines such as CellTFusion through a user-provided fold construction function.

Usage

wrapper_train_best_hyperparams_survival(
  train_data,
  optimized,
  ml_method,
  fold_construction_fun,
  fold_construction_args_fixed,
  outcome_col = "time",
  event_col = "event"
)

Arguments

train_data

A data frame containing the original training data used for cross-validation.

optimized

A list output from aggregate_results_survival() or compute_k_fold_CV_survival(), containing the best-tuned parameters (bestTune) and model performance summaries.

ml_method

Character string specifying the survival model to train. Must be one of:

• "cox_ph_survival" — Cox proportional hazards model.

• "proportional_hazards_glmnet" — Penalized Cox (elastic net).

• "survreg_flexsurv" — Parametric AFT model.

• "rand_forest_partykit" — Random survival forest via partykit.

• "rand_forest_aorsf" — Oblique random survival forest.

• "decision_tree_partykit" — Single survival tree.

• "bag_tree_rpart" — Bagged CART-based survival trees.

• "boost_tree_mboost" — Gradient boosting for censored data.

fold_construction_fun

A custom function used to construct folds and preprocessed data (e.g., prepare_CellTFusion_folds()). Must accept arguments data and optionally bestune.

fold_construction_args_fixed

A named list of fixed arguments to pass to fold_construction_fun() (e.g., paths, deconvolution matrices, etc.).

outcome_col

Character string naming the survival time column (default = "time").

event_col

Character string naming the event indicator column (default = "event").

Value

A named list containing:

Model

A list containing the fitted parsnip model object, resampling results, and tuning information.

training_set

The final preprocessed training dataset used for fitting.

custom_output

Additional data returned by the custom fold construction function (e.g., CellTFusion outputs or parameter tables).

Details

This function performs the following steps:

1. Extracts the optimal hyperparameters from the optimized object.

2. Reconstructs the training dataset using the provided fold_construction_fun(), including any custom preprocessing or feature generation.

3. Applies the optimal hyperparameters to the model specification.

4. Fits the final model using the full training data.

If the selected model type has no tunable hyperparameters, the function automatically detects this and proceeds with the default model configuration.

See also

compute_k_fold_CV_survival(), aggregate_results_survival(), compute_ml_survival()