Skip to contents

Parameter Reference Guide

Source: vignettes/parameter-reference.Rmd
parameter-reference.Rmd

E2E Package Parameter Reference Guide

This guide provides comprehensive parameter documentation for all E2E functions.

Built-in Datasets

E2E includes example datasets for both diagnostic and prognostic modeling:

Diagnostic Datasets

  • train_dia: Training data with sample IDs (column 1), outcomes 0/1 (column 2), and features (columns 3+)
  • test_dia: Test data with the same structure

Prognostic Datasets

  • train_pro: Training data with sample IDs (column 1), survival status 0/1 (column 2), survival time (column 3), and features (columns 4+)
  • test_pro: Test data with the same structure

Built-in Models

Diagnostic Models (12 algorithms)

  • rf: Random Forest
  • xb: XGBoost
  • svm: Support Vector Machine
  • mlp: Multi-Layer Perceptron
  • lasso: L1-regularized Logistic Regression
  • en: Elastic Net
  • ridge: L2-regularized Logistic Regression
  • lda: Linear Discriminant Analysis
  • qda: Quadratic Discriminant Analysis
  • nb: Naive Bayes
  • dt: Decision Tree
  • gbm: Gradient Boosting Machine

Prognostic Models (6 algorithms)

  • lasso_pro: Lasso Cox Regression
  • en_pro: Elastic Net Cox Regression
  • ridge_pro: Ridge Cox Regression
  • stepcox_pro: Stepwise Cox Regression
  • gbm_pro: Gradient Boosting Machine
  • rsf_pro: Random Survival Forest

Diagnostic Modeling Functions

models_dia()

Trains base classification models for diagnostic tasks. Parameters:

  • data (required): Data frame with sample names (column 1), outcomes 0/1 (column 2), features (columns 3+)

  • model (required): Character vector of model names or “all_dia” for all models

  • tune: Logical (default FALSE). Whether to perform hyperparameter tuning

  • threshold_choices: Threshold selection method

    • “default” (default): Fixed 0.5 threshold
    • “f1”: Optimize F1 score
    • “youden”: Optimize Youden index
    • Numeric value (0-1): Custom threshold

  • seed: Integer (default 123). Random seed for reproducibility

bagging_dia()

Bootstrap aggregating ensemble method. Parameters:

  • data (required): Training data frame

  • base_model_name (required): Base model name (e.g., “xb”, “rf”)

  • n_estimators: Integer (default 50). Number of base models

  • subset_fraction: Numeric (default 0.632). Bootstrap sampling fraction

  • tune_base_model: Logical (default FALSE). Tune base models

  • threshold_choices: Same as models_dia()

  • seed: Integer (default 123). Random seed

voting_dia()

Voting ensemble combining multiple models. Parameters:

  • results_all_models (required): Output from models_dia()

  • data (required): Training data

  • type: Voting type

    • “soft” (default): Weighted probability averaging
    • “hard”: Majority voting

  • weight_metric: String (default “AUROC”). Metric for soft voting weights

  • top: Integer (default 5). Number of top models to use

  • threshold_choices: Same as models_dia()

  • seed: Integer (default 123). Random seed

stacking_dia()

Stacking ensemble with meta-model. Parameters:

  • results_all_models (required): Output from models_dia()

  • data (required): Training data

  • meta_model_name (required): Meta-model name (e.g., “lasso”, “gbm”)

  • top: Integer (default 5). Number of top base models

  • tune_meta: Logical (default FALSE). Tune meta-model

  • threshold_choices: Same as models_dia()

  • seed: Integer (default 123). Random seed

imbalance_dia()

Handles imbalanced datasets using EasyEnsemble-like algorithm. Parameters:

  • data (required): Imbalanced training data

  • base_model_name (required): Base model for balanced subsets

  • n_estimators: Integer (default 10). Number of balanced subsets

  • tune_base_model: Logical (default FALSE). Tune base models

  • threshold_choices: Same as models_dia()

  • seed: Integer (default 123). Random seed

apply_dia()

Applies trained model to new data. Parameters:

  • trained_model_object (required): Trained model object from E2E functions

  • new_data (required): New data for prediction (sample IDs in column 1)

  • label_col_name: String (default NULL). True label column name if available

evaluate_predictions_dia()

Evaluates model predictions. Parameters:

  • prediction_df (required): Prediction data frame from apply_dia()

  • threshold_choices: Same as models_dia()

Prognostic Modeling Functions

models_pro()

Trains base survival models. Parameters:

  • data (required): Data frame with sample ID, survival status, time, features

  • model (required): Model names or “all_pro” for all models

  • tune: Logical (default FALSE). Hyperparameter tuning

  • time_unit: String (default “day”). Time unit (“day”, “month”, “year”)

  • years_to_evaluate: Numeric vector (default c(1,3,5)). Time points for time-dependent AUROC

  • seed: Integer (default 789). Random seed

stacking_pro()

Stacking ensemble for survival analysis. Parameters:

  • results_all_models (required): Output from models_pro()

  • data (required): Training data

  • meta_model_name (required): Meta-model name

  • top: Integer (default 3). Number of top base models

  • tune_meta: Logical (default FALSE). Tune meta-model

  • time_unit: String (default “day”). Time unit

  • years_to_evaluate: Numeric vector (default c(1,3,5)). Evaluation time points

  • seed: Integer (default 789). Random seed

bagging_pro()

Bootstrap aggregating for survival analysis. Parameters:

  • data (required): Training data

  • base_model_name (required): Base model name

  • n_estimators: Integer (default 10). Number of base models

  • subset_fraction: Numeric (default 0.632). Bootstrap sampling fraction

  • tune_base_model: Logical (default FALSE). Tune base models

  • time_unit: String (default “day”). Time unit

  • years_to_evaluate: Numeric vector (default c(1,3,5)). Evaluation time points

  • seed: Integer (default 456). Random seed

apply_pro()

Applies trained survival model to new data. Parameters:

  • trained_model_object (required): Trained model object

  • new_data (required): New data with same structure as training data

  • time_unit: String (default “day”). Time unit

evaluate_predictions_pro()

Evaluates survival model predictions. Parameters:

  • prediction_df (required): Prediction data frame from apply_pro()

  • years_to_evaluate: Numeric vector (default c(1,3,5)). Evaluation time points

Visualization Functions

figure_dia()

Creates diagnostic model evaluation plots. Parameters:

  • type (required): Plot type
    • “roc”: ROC curve
    • “prc”: Precision-recall curve
    • “matrix”: Confusion matrix
  • data (required): Model results object

figure_pro()

Creates prognostic model evaluation plots. Parameters:

  • type (required): Plot type

    • “km”: Kaplan-Meier survival curves
    • “tdroc”: Time-dependent ROC curves

  • data (required): Model results object

  • time_unit: String (default “days”). Time unit for axis labels

figure_shap()

Creates SHAP interpretation plots. Parameters:

  • data (required): Model results with sample_score data frame

  • raw_data (required): Original feature data

  • target_type (required): Data type

    • “diagnosis”: Features start from column 3
    • “prognosis”: Features start from column 4

Custom Model Registration

register_model_dia() / register_model_pro()

Registers custom algorithms.

Usage: 1. Define custom function following E2E conventions 2. Register with register_model_dia("model_name", custom_function) 3. Use registered model in E2E workflows