mxPenalty {OpenMx}

This function creates a penalty object

Description

This function creates a penalty object

Usage

mxPenalty(
  what,
  epsilon = 1e-05,
  scale = 1,
  how = imxPenaltyTypes,
  smoothProportion = 0.05,
  hyperparams = c(),
  hpranges = list(),
  name = NULL
)

Arguments

what

A character vector of parameters to regularize

epsilon

how close to zero is zero?

scale

a given parameter is divided by scale before comparison with epsilon

how

what kind of function to use

smoothProportion

what proportion of the region between epsilon and zero should be used to smooth the penalty function

hyperparams

a character vector of hyperparameter names

hpranges

a named list of hyperparameter ranges. Used in search if no ranges are specified.

name

Name of the regularizer object

Details

mxPenalty expects to find an mxMatrix with free parameters that correspond to all named hyperparameters.

Gradient descent optimizers are designed for and work best on smooth functions. All of the regularization penalties implemented traditionally contain discontinuities. By default, OpenMx uses smoothed versions of these functions. Smoothing is controlled by smoothProportion. If smoothProportion is zero then the traditional discontinuous functions are used. Otherwise, smoothProportion of the region between epsilon and zero is used for smoothing.