VariableImportance: Computes Exact or Approximate L0-penalization based Variable...

In BranchGLM: Efficient Best Subset Selection for GLMs via Branch and Bound Algorithms

Computes Exact or Approximate L0-penalization based Variable Importance for GLMs

Description

Gets exact or approximate L0-penalization based variable importance values for generalized linear models (GLMs). More details about what the variable importance values are can be found in the details section.

Usage

VariableImportance(
  object,
  VIMethod = "simultaneous",
  parallel = FALSE,
  nthreads = 8,
  showprogress = TRUE
)

Arguments

object	an object of class BranchGLMVS.
VIMethod	one of "separate" or "simultaneous" to denote the method used to find the variable importance values. This is ignored if the type of variable selection employed in object was a heuristic method.
parallel	a logical value to indicate if parallelization should be used.
nthreads	number of threads used with OpenMP, only used if parallel = TRUE.
showprogress	a logical value to indicate whether or not to show progress updates.

Details

Note that variable importance values can only be found for sets of variables that are not kept through the model selection process. More details about the variable importance values will be made available in an upcoming paper.

When a branch and bound algorithm is used in object, then the exact variable importance values are computed. When a heuristic method is used, then approximate variable importance values are computed based on the specified heuristic method.

Value

A BranchGLMVI object which is a list with the following components

results	a data.frame with the variable importance values and degrees of freedom
metric	metric used to select the best models
numchecked	number of models fit
VS	the supplied BranchGLMVS object
with	a numeric matrix with the best models that include each set of variables
withmetrics	a numeric vector with the metric values for the best models with each set of variables
without	a numeric matrix with the best models that exclude each set of variables
withoutmetrics	a numeric vector with the metric values for the best models without each set of variables

References

Seedorff J, Cavanaugh JE. Assessing Variable Importance for Best Subset Selection. Entropy. 2024; 26(9):801. \Sexpr[results=rd]{tools:::Rd_expr_doi("https://doi.org/10.3390/e26090801")}

See Also

VariableImportance.boot, barplot.BranchGLMVI

Examples

Data <- iris
Fit <- BranchGLM(Sepal.Length ~ ., data = Data, family = "gaussian", link = "identity")

# Doing branch and bound selection 
VS <- VariableSelection(Fit, type = "branch and bound", metric = "BIC", 
showprogress = FALSE)

# Getting variable importance
VI <- VariableImportance(VS, showprogress = FALSE)
VI

# Plotting variable importance
oldmar <- par("mar")
par(mar = c(4, 6, 3, 1) + 0.1)
barplot(VI)
par(mar = oldmar)

BranchGLM documentation built on Sept. 28, 2024, 9:07 a.m.