library(E2E)
# Initialize systems for the examples
initialize_modeling_system_dia()
#> Diagnostic modeling system initialized and default models registered.
initialize_modeling_system_pro()
#> Prognosis modeling system initialized.This guide covers advanced topics such as extending the E2E framework with custom models and interpreting model predictions using SHAP.
The E2E framework is fully extensible. You can register your own custom models. For example, to add Adaboost for diagnosis:
# 1. Define the model function (must accept X, y, and other standard args)
ab_dia <- function(X, y, tune = FALSE, cv_folds = 5) {
# Ensure caret is available
if (!requireNamespace("caret", quietly = TRUE)) {
stop("Package 'caret' is required for this custom model.")
}
ctrl <- caret::trainControl(method = "cv", number = cv_folds,
classProbs = TRUE, summaryFunction = caret::twoClassSummary)
grid <- if (tune) {
expand.grid(iter = c(50, 100), maxdepth = c(1, 2), nu = 0.1)
} else {
expand.grid(iter = 50, maxdepth = 1, nu = 0.1)
}
caret::train(x = X, y = y, method = "ada", metric = "ROC", trControl = ctrl, tuneGrid = grid)
}
# 2. Register the model with a unique name
register_model_dia("ab", ab_dia)
# 3. Now you can use "ab" in any diagnostic function
results_ab <- models_dia(train_dia, model = "ab")
#> Running model: ab
print_model_summary_dia("ab", results_ab$ab)
#>
#> --- ab Model (on Training Data) Metrics ---
#> Threshold Strategy: default (0.5000)
#> AUROC: 0.9994 (95% CI: 0.9987 - 1.0000)
#> AUPRC: 0.9999
#> Accuracy: 0.9919
#> F1: 0.9955
#> Precision: 0.9936
#> Recall: 0.9974
#> Specificity: 0.9375
#> --------------------------------------------------# and also you can add coxboost as a prognosis model like
#' @title Train CoxBoost
#your_model_pro <- function(X, y_surv, tune = FALSE) {
# if (!requireNamespace("CoxBoost", quietly = TRUE)) {
# stop("Package 'CoxBoost' is required for this function but is not installed.
# Please install it from GitHub using: remotes::install_github('binderh/CoxBoost')")}
# X_matrix <- as.matrix(X)
# stepno_val <- 100
# if (tune) stepno_val <- 100
# fit <- CoxBoost::CoxBoost(time = y_surv[,1], status = y_surv[,2], x = X_matrix, stepno = stepno_val, penalty = 100)
# fit$fitted_scores <- as.vector(predict(fit, newdata = X_matrix, type = "lp"))
# structure(
# list(finalModel = fit, X_train_cols = colnames(X), model_type = "survival_coxboost"),
# class = c("survival_coxboost", "pro_model")
# )
#}
#
#' @export
#predict_pro.survival_coxboost <- function(object, newdata, ...) {
# if (!requireNamespace("CoxBoost", quietly = TRUE)) {
# stop("Package 'CoxBoost' is needed for prediction.")
# }
# newdata <- .ensure_features(object, newdata)
# X_matrix <- as.matrix(newdata)
# as.vector(predict(object$finalModel, newdata = X_matrix, type = "lp"))
#}
#register_model_pro("cb_pro", cb_pro)figure_shap)The figure_shap function provides model-agnostic
explanations by calculating SHAP values. This reveals which features had
the most impact on the model’s output.
# First, we need a model result object
bagging_xb_results <- bagging_dia(train_dia, base_model_name = "xb", n_estimators = 10, seed=123)
#> Running Bagging model: Bagging_dia (base: xb)
# Now, generate the SHAP explanation plot
p6 <- figure_shap(
data = bagging_xb_results,
raw_data = train_dia,
target_type = "diagnosis"
)
#> Training 'xgboost' surrogate model and calculating SHAP values...
#plot(p6)# First, we need a model result object
stacking_stepcox_pro_results <- stacking_pro(
results_all_models = models_pro(train_pro, model = c("lasso_pro", "rsf_pro")),
data = train_pro,
meta_model_name = "stepcox_pro"
)
#> Running model: lasso_pro
#> Running model: rsf_pro
# Generate the SHAP explanation plot
p7 <- figure_shap(
data = stacking_stepcox_pro_results,
raw_data = train_pro,
target_type = "prognosis"
)
#> Training 'xgboost' surrogate model and calculating SHAP values...
#plot(p7)