Tue, 10/06/2009 - 17:54

There's been some discussion among the beta testers to support setting a free parameter to a fixed parameter across an entire model, or setting a fixed parameter to a free parameter across an entire model. Here's a function in R that will accomplish the transformation. The 'model argument is a MxModel object. The 'labels' argument is a vector of parameter names. The 'free' argument is either a NA or a single TRUE or FALSE. The 'value' argument is either a NA or a single numeric value. If 'free' is not NA, then all the parameters with name 'labels' will have their free/fixed status changed to 'free'. If 'value' is not NA, then all the parameters with name 'labels' with have their starting values changed to 'value'.

setParameters <- function(model, labels, free = NA, value = NA) { if(missing(model) || !is(model, "MxModel")) { stop("The 'model' argument must be a MxModel object") } if(missing(labels) || typeof(labels) != "character") { stop("The 'labels' argument must be a vector of characters") } if(!is.na(free) && (typeof(free) != "logical" || length(free) != 1)) { stop("The 'free' argument must be a single boolean value") } if(!is.na(value) && (!is.numeric(value) || length(value) != 1)) { stop("The 'value' argument must be a single numeric value") } return(setParametersHelper(model, labels, free, value)) } setParametersHelper <- function(model, labels, free, value) { model@matrices <- lapply(model@matrices, setParametersMatrix, labels, free, value) model@submodels <- lapply(model@submodels, setParametersHelper, labels, free, value) return(model) } setParametersMatrix <- function(matrix, labels, free, value) { select <- apply(matrix@labels, c(1,2), function(x) {!is.na(x) && x %in% labels}) if (!is.na(free)) { matrix@free[select] <- free } if (!is.na(value)) { matrix@values[select] <- value } return(matrix) }

Is this implemented in the latest build? Do you have an example?

Thanks!

The functions are not part of OpenMx, so to implement it, you would simply paste the function code into your demo scripts: then it is available to be called.

I added an example using this (univariate ACE, dropping c by label)

trunk/models/passing/univACE_drop_helper.R

It would be great if this type of syntax worked

model = setParameters(model, labels=c("share.c1"), free = FALSE, value = 0)

i.e., pointing into a sub-model

also this highlights how helpful this parameter of summary() (or perhaps mxRun) would be

summary(fit, sat=xxx)

then the AIC would be computed between the old and new models.

setParameters() recursively transforms all the parameters in the model hierarchy. To transform only a sub-tree the following would work:

`model$share <- setParameters(model$share, "c1", FALSE, 0)`

To write a non-recursive version of setParameters, delete the second line in the setParametersHelper function.