penaltyLearning (version 2020.5.13)

IntervalRegressionCVmargin: IntervalRegressionCVmargin

Description

Use cross-validation to fit an L1-regularized linear interval regression model by optimizing both margin and regularization parameters. This function just calls IntervalRegressionCV with a margin.vec parameter that is computed based on the finite target interval limits. If default parameters are used, this function should be about 10 times slower than IntervalRegressionCV (since this function computes n.margin=10 models per regularization parameter whereas IntervalRegressionCV only computes one). On large (N > 1000 rows) data sets, this function should yield a model which is a little more accurate than IntervalRegressionCV (since the margin parameter is optimized).

Usage

IntervalRegressionCVmargin(feature.mat, 
    target.mat, log10.diff = 2, 
    n.margin = 10L, ...)

Arguments

feature.mat

Numeric feature matrix, n observations x p features.

target.mat

Numeric target matrix, n observations x 2 limits.

log10.diff

Numeric scalar: factors of 10 below the largest finite limit difference to use as a minimum margin value (difference on the log10 scale which is used to generate margin parameters). Bigger values mean a grid of margin parameters with a larger range. For example if the largest finite limit in target.mat is 26 and the smallest finite limit is -4 then the largest limit difference is 30, which will be used as the maximum margin parameter. If log10.diff is the default of 2 then that means the smallest margin parameter will be 0.3 (two factors of 10 smaller than 30).

n.margin

Integer scalar: number of margin parameters, by default 10.

…

Passed to IntervalRegressionCV.

Value

Model fit list from IntervalRegressionCV.

Examples

# NOT RUN {
if(interactive()){
  library(penaltyLearning)
  data(
    "neuroblastomaProcessed",
    package="penaltyLearning",
    envir=environment())
  if(require(future)){
    plan(multiprocess)
  }
  set.seed(1)
  fit <- with(neuroblastomaProcessed, IntervalRegressionCVmargin(
    feature.mat, target.mat, verbose=1))
  plot(fit)
  print(fit$plot.heatmap)
}
# }