mxPenaltySearchExternal: External search for an optimal penalty value

In trbrick/mxregsem: Regularization of models in OpenMx

Description

Perform a grid search over a space of regularization penalties. This happens in the front-end, returning to R at each cycle. Currently only guaranteed to work for single-parameter models. Expect many changes in the interface while this is developed further. **WARNING: Dangerously assumes that fitfunction is the -2 log likelihood**

Usage

mxPenaltySearchExternal(
  model,
  search_space = NULL,
  epsilon = 1e-06,
  ...,
  approach = "EBIC",
  ebicGamma = 0.5,
  returnConstrained = FALSE,
  verbose = interactive()
)

Arguments

model	an MxRegularizedModel object to regularize
epsilon	how close to zero is zero?
...	the search space for this sequence in format *param=c(...)* where c(...) is the list of possible values for parameter param.
approach	what fit function to use? Currently only EBIC is available
ebicGamma	what Gamma value to use for EBIC? Must be between 0 and 1
returnConstrained	fix all regularized-to-zero values to zero, and rerun fit without regularization

Details

The EBIC computation uses the computation from the previous regsem. It is not clear why this is the case.

TODO: Figure out how to return all the individual runs.

Value

A regularized MxModel; at this point an MxRegularizedModel

trbrick/mxregsem documentation built on Nov. 18, 2020, 7:30 p.m.