softening.optimized {SplitSoftening}

Make split softening optimized with Nelder-Mead.

Description

This softening configures all parameters in the tree with optimization method Nelder-Mead to minimize the given ‘miss’ function.

Usage

softening.optimized(
  tr,
  d,
  miss.fn,
  verbosity = 0,
  implementation = c("gsl", "R"),
  iteration.count = NULL,
  sft.ini = 1
)

Arguments

tr

The soft tree

d

The data set to be used in intialization for determining data boundaries and in optimization step to evaluate the objective function on the predictions on this data set by the soft tree with updated softening parameters.

miss.fn

Function to provide the value of the objective function for optimization.

The function obtains as an argument the matrix of class probabilities as returned by predictSoftsplits when making predictions for the data set d using the soft tree tr but with some softening parameters reset within optimization procedure. The function is expected to return one numeric value; this value is minimized by the optimization method.

verbosity

The verbosity level configures how many additional information is printed

implementation

Indentify implementation of optimizer.

iteration.count

Number of optimizer iterations.

sft.ini

Parameter of softening used as the initial value for the optimization.

• "gsl" uses multimin function from gsl package. Note: In the current version (2.1-6) of gsl package this function does not work.

• "R" uses optim - the standard optimization function in R.

Value

The soft tree with the new softening parameters