IntervalRegressionRegularized {penaltyLearning}

IntervalRegressionRegularized

Description

Repeatedly use IntervalRegressionInternal to solve interval regression problems for a path of regularization parameters. This function does not perform automatic selection of the regularization parameter; instead, it returns regression models for a range of regularization parameters, and it is up to you to select which one to use. For automatic regularization parameter selection, use IntervalRegressionCV.

Usage

IntervalRegressionRegularized(feature.mat, 
    target.mat, initial.regularization = 0.001, 
    factor.regularization = 1.2, 
    verbose = 0, margin = 1, 
    ...)

Arguments

feature.mat

Numeric feature matrix.

target.mat

Numeric target matrix.

initial.regularization

Initial regularization parameter.

factor.regularization

Increase regularization by this factor after finding an optimal solution. Or NULL to compute just one model (initial.regularization).

verbose

Print messages if >= 1.

margin

Non-negative margin size parameter, default 1.

...

Other parameters to pass to IntervalRegressionInternal.

Value

List representing fit model. You can do fit$predict(feature.matrix) to get a matrix of predicted log penalty values. The param.mat is the n.features * n.regularization numeric matrix of optimal coefficients (on the original scale).

Author(s)

Toby Dylan Hocking

Examples

if(interactive()){
  library(penaltyLearning)
  data("neuroblastomaProcessed", package="penaltyLearning", envir=environment())
  i.train <- 1:500
  fit <- with(neuroblastomaProcessed, IntervalRegressionRegularized(
    feature.mat[i.train,], target.mat[i.train,]))
  plot(fit)
}