mxFitFunctionML: Create MxFitFunctionML Object

Description

This function creates a new MxFitFunctionML object.

Usage

mxFitFunctionML(vector = FALSE, rowDiagnostics = FALSE, ..., fellner =
  as.logical(NA), verbose=0L, profileOut=c(),
  rowwiseParallel=as.logical(NA), jointConditionOn = c("auto", "ordinal", "continuous"))

Arguments

vector

A logical value indicating whether the objective function result is the likelihood vector.

rowDiagnostics

A logical value indicating whether the row-wise results of the objective function should be returned as an attribute of the fit function.

...

Not used. Forces remaining arguments to be specified by name.

fellner

Whether to fully expand the covariance matrix for maximum flexibility.

verbose

Level of diagnostic output

profileOut

Character vector naming constant coefficients to profile out of the likelihood (sometimes known as REML)

rowwiseParallel

For raw data only, whether to use OpenMP to parallelize the evaluation of rows

jointConditionOn

The evaluation strategy when both continuous and ordinal data are present.

Value

Returns a new MxFitFunctionML object. One and only one MxFitFunctionML object should be included in each model along with an associated mxExpectationNormal or mxExpectationRAM object.

Details

Fit functions are functions for which free parameter values are optimized such that the value of a cost function is minimized. The mxFitFunctionML function computes -2*(log likelihood) of the data given the current values of the free parameters and the expectation function (e.g., mxExpectationNormal or mxExpectationRAM) selected for the model.

The 'vector' argument is either TRUE or FALSE, and determines whether the objective function returns a column vector of the likelihoods, or a single -2*(log likelihood) value.

The 'rowDiagnostics' argument is either TRUE or FALSE, and determines whether the row likelihoods are returned as an attribute of the fit function. Additionally, the squared Mahalanobis distance and the number of observed (non-missing) variables) for each row are returned under the names rowDist and rowObs, respectively. It is sometimes useful to inspect the likelihoods for outliers, diagnostics, or other anomalies. Each rowwise squared Mahalanobis distance should be chi-squared distributed with degrees of freedom equal to the number of observed variables. In the case of no missing data, all of the rowwise squared Mahalanobis distances should theoretically be chi-squared distributed with the same degrees of freedom. In the case of some missing data, the rowwise squared Mahalanobis distances should theoretically be a mixture of chi-squared distributions with mixing proportions equal to the proportions of each number of observed variables.

If there are ordinal data, then only the row likelihoods are returned among the row diagnostics.

When vector=FALSE and rowDiagnostics=TRUE, the fit function can be referenced in the model and included in algebras as a scalar. The row likelihoods, row distances, and row observations are then an attribute of the fit function but are not accessible in the model during optimization. The row likelihoods and other diagnostics are accessible to the user after the model has been run.

By default, jointConditionOn='auto' and a heuristic will be used to select the fastest algorithm. Conditioning the continuous data on ordinal will be superior when there are relatively few unique ordinal patterns. Otherwise, conditioning the ordinal data on continuous will perform better when there are relatively many ordinal patterns.

Usage Notes:

The results of the optimization can be reported using the summary function, or accessed directly in the 'output' slot of the resulting model (i.e., modelName$output). Components of the output may be referenced using the Extract functionality.

References

The OpenMx User's guide can be found at http://openmx.ssri.psu.edu/documentation.

Examples

Run this code

# NOT RUN {
# Create and fit a model using mxMatrix, mxAlgebra, mxExpectationNormal, and mxFitFunctionML

library(OpenMx)

# Simulate some data

x=rnorm(1000, mean=0, sd=1)
y= 0.5*x + rnorm(1000, mean=0, sd=1)
tmpFrame <- data.frame(x, y)
tmpNames <- names(tmpFrame)

# Define the matrices

M <- mxMatrix(type = "Full", nrow = 1, ncol = 2, values=c(0,0), 
              free=c(TRUE,TRUE), labels=c("Mx", "My"), name = "M")
S <- mxMatrix(type = "Full", nrow = 2, ncol = 2, values=c(1,0,0,1), 
              free=c(TRUE,FALSE,FALSE,TRUE), labels=c("Vx", NA, NA, "Vy"), name = "S")
A <- mxMatrix(type = "Full", nrow = 2, ncol = 2, values=c(0,1,0,0), 
              free=c(FALSE,TRUE,FALSE,FALSE), labels=c(NA, "b", NA, NA), name = "A")
I <- mxMatrix(type="Iden", nrow=2, ncol=2, name="I")

# Define the expectation

expCov <- mxAlgebra(solve(I-A) %*% S %*% t(solve(I-A)), name="expCov")
expFunction <- mxExpectationNormal(covariance="expCov", means="M", dimnames=tmpNames)

# Choose a fit function

fitFunction <- mxFitFunctionML(rowDiagnostics=TRUE)
# also return row likelihoods, even though the fit function
#  value is still 1x1

# Define the model

tmpModel <- mxModel(model="exampleModel", M, S, A, I, expCov, expFunction, fitFunction, 
                    mxData(observed=tmpFrame, type="raw"))

# Fit the model and print a summary

tmpModelOut <- mxRun(tmpModel)
summary(tmpModelOut)

fitResOnly <- mxEval(fitfunction, tmpModelOut)
attributes(fitResOnly) <- NULL
fitResOnly

# Look at the row likelihoods alone
fitLikeOnly <- attr(mxEval(fitfunction, tmpModelOut), 'likelihoods')
head(fitLikeOnly)

# }

Run the code above in your browser using DataCamp Workspace