GeDS (version 0.3.3)

bl_imp.GeDSboost: Base Learner Importance for GeDSboost Objects

Description

S3 method for "GeDSboost" class objects that calculates the in-bag risk reduction ascribable to each base-learner of an FGB-GeDS model. Essentially, it measures and aggregates the decrease in the empirical risk attributable to each base-learner for every time it is selected across the boosting iterations. This provides a measure on how much each base-learner contributes to the overall improvement in the model's accuracy, as reflected by the decrease in the empiral risk (loss function). This function is adapted from varimp and is compatible with the available mboost-package methods for varimp, including plot, print and as.data.frame.

Usage

# S3 method for GeDSboost
bl_imp(object, boosting_iter_only = FALSE, ...)

Value

An object of class varimp with available plot, print and as.data.frame methods.

Arguments

object

An object of class "GeDSboost" class.

boosting_iter_only

Logical value, if TRUE then base-learner in-bag risk reduction is only computed across boosting iterations, i.e., without taking into account a potential initial GeDS learner.

...

Potentially further arguments.

Details

See varimp for details.

References

Hothorn T., Buehlmann P., Kneib T., Schmid M. and Hofner B. (2022). mboost: Model-Based Boosting. R package version 2.9-7, https://CRAN.R-project.org/package=mboost.

Examples

library(GeDS)
library(TH.data)

data("bodyfat", package = "TH.data")
N <- nrow(bodyfat); ratio <- 0.8
set.seed(123)
trainIndex <- sample(1:N, size = floor(ratio * N))
# Subset the data into training and test sets
train <- bodyfat[trainIndex, ]
test <- bodyfat[-trainIndex, ]

Gmodboost <- NGeDSboost(formula = DEXfat ~ f(hipcirc) + f(kneebreadth) + f(anthro3a),
                        data = train, phi = 0.7, initial_learner = FALSE)
MSE_Gmodboost_linear <- mean((test$DEXfat - predict(Gmodboost, newdata = test, n = 2))^2)
MSE_Gmodboost_quadratic <- mean((test$DEXfat - predict(Gmodboost, newdata = test, n = 3))^2)
MSE_Gmodboost_cubic <- mean((test$DEXfat - predict(Gmodboost, newdata = test, n = 4))^2)

# Print MSE
cat("\n", "TEST MEAN SQUARED ERROR", "\n",
    "Linear NGeDSboost:", MSE_Gmodboost_linear, "\n",
    "Quadratic NGeDSboost:", MSE_Gmodboost_quadratic, "\n",
    "Cubic NGeDSboost:", MSE_Gmodboost_cubic, "\n")

# Base Learner Importance
bl_imp <- bl_imp(Gmodboost)
print(bl_imp)
plot(bl_imp)