MultiCoreOptimRegModel: MultiCoreOptimRegModel

In jhorzek/regmx:

MultiCoreOptimRegModel

Description

Note: regmx is based on the R package regsem. Because of the early status of regmx, it is recommended to use regsem instead! MultiCoreOptimRegModel creates a range of regularized models from an mxModel. It automatically tests different penalty values and returns the best model. It only uses multiple cores.

Usage

MultiCoreOptimRegModel(
  mxModelObject,
  alpha = 1,
  gamma = 0,
  regOn,
  regIndicators,
  regValues,
  criterion = "BIC",
  autoCV = FALSE,
  k = 5,
  Boot = FALSE,
  manualCV = NULL,
  zeroThresh = 0.001,
  scaleCV = TRUE,
  cores = 1
)

Arguments

mxModelObject	an already run mxModel
alpha	alpha controls the type of penalty. For lasso regularization, set alpha = 1, for ridge alpha = 0. Values between 0 and 1 implement elastic net regularization
gamma	gamma sets the power in the denominator of parameter specific weights when using adaptive lasso regularization. Make sure to set alpha to 1 when using a gamma other than 0.
regOn	string vector with matrices that should be regularized. The matrices must have the same name as the ones provided in the mxModelObject (e.g., "A")
regIndicators	list of matrices indicating which parameters to regularize in the matrices provided in regOn. The matrices in regIndicators must to have the same names as the matrices they correspond to (e.g., regIndicators = list("A" = diag(10))). 1 Indicates a parameter that will be regularized, 0 an unregularized parameter
criterion	Criterion for chosing the best final model. Possible are: AIC, BIC, CV.m2LL (only if manualCV is provided)
autoCV	logical indicating if cross-validation should be performed automatically
k	number of splits performed when autoCV = TRUE
Boot	logical indicating if Bootstrap should be performed. Not yet implemented!
manualCV	if cross-validation should be performed manually, provide a cross-validation sample (has to be of the same class as the data in the mxModelObject; e.g., mxData)
zeroThresh	Threshold at which parameters are evaluated as being zero (necessary for AIC and BIC)
scaleCV	indicate if the CV samples should be scaled automatically
regValue	numeric value depicting the penalty size
regValue_start	initial penalty value (recommended: 0)
regValue_end	highest penalty value tested
regValue_stepsize	increase in penValue between iterations

Author(s)

Jannik Orzek

Examples

# The following example is adapted from the regsem help to demonstrate the equivalence of both methods:

library(lavaan)
library(OpenMx)
# put variables on same scale for regsem
HS <- data.frame(scale(HolzingerSwineford1939[,7:15]))

# define variables:
latent = c("f1")
manifest = c("x1","x2","x3","x4","x5", "x6", "x7", "x8", "x9")

# define paths:
loadings <- mxPath(from = latent, to = manifest, free = c(F,T,T,T,T,T,T,T,T), values = 1)
lcov <- mxPath(from = latent, arrows = 2, free = T, values = 1)
lmanif <- mxPath(from = manifest, arrows =2 , free =T, values = 1)

# define model:
myModel <- mxModel(name = "myModel", latentVars = latent, manifestVars = manifest, type = "RAM",
                   mxData(observed = HS, type = "raw"), loadings, lcov, lmanif,
                   mxPath(from = "one", to = manifest, free = T)
)

fit_myModel <- mxRun(myModel)

# Show the names of the matrices in the model:
names(fit_myModel$matrices)

# Penalize specific parameters from the A matrix (directional paths):
regOn <- c("A")

selectedA <- matrix(0, ncol = ncol(fit_myModel$A$values), nrow = nrow(fit_myModel$A$values))
selectedA[c(2,3,7,8,9),10] <-1 # parameters that should be regularized have to be marked with 1
regIndicators <- list("A" = selectedA) # save in a list. Note the naming of the list element

# Run the models:

reg_model <- optimRegModel(mxModelObject = fit_myModel, alpha = 1, gamma = 0, regOn  = regOn, regIndicators = regIndicators, cores = 2)

reg_model$`fit measures`

reg_model$`best penalty`

# Run the same model with 5-fold cross-validation

CV_reg_model <- optimRegModel(mxModelObject = fit_myModel, alpha = 1, gamma = 0 regOn  = regOn, regIndicators = regIndicators,
                              autoCV = T, k = 5, cores = 2)
CV_reg_model$`CV results`

jhorzek/regmx documentation built on Sept. 19, 2022, 2:30 a.m.