How to add object to a model without re-run

Posted on Mon, 07/15/2024 - 12:22

JuanJMV Joined: 07/20/2016

Forums

Dear all,

I have fitted a very computationally demanding model (around one week) and now I need to add an mxAlgebra object into this model:

threG2 <- mxAlgebra( expression= threG[,1:7], name="threG2" )

I know I can do it in this way:

modelIPAE2 <- mxModel( fitIPAE,threG2, name="mulIPAE2c" ) fitIPAE2 <- mxTryHardOrdinal( modelIPAE2, intervals=F, extraTries=11 )

But I wonder if there is any way to modify the model without re-running it

I need it because I would like to use this line later:

fsModel@expectation <- mxExpectationNormal("as","mean2","threG[,1:7]",dimnames=selVars[1:7])

But I cannot make a subset of a S4 object, so I need to use this:

fsModel@expectation <- mxExpectationNormal("as","mean2","threG2",dimnames=selVars[1:7])

But I previously need to have threG2 in my model.

Thank you so much in advance!

Replied on Tue, 07/23/2024 - 11:09

AdminRobK Joined: 01/24/2014

more info?

Could you say a little bit more about what you're trying to do here? A full script (or at least part of one) could help provide some context.

Replied on Tue, 07/23/2024 - 14:27

JuanJMV Joined: 07/20/2016

Hi Rob,

Thank you so much for your help.

Attached you can find the script that I am using.

First you can find a multivariate model including an independent pathway model (Boulder's script) (lines 1 to 190).

Then I have a script (I adapted it from this paper (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3773873/) to calculate individual factor scores for A an E (lines 191-230).

It has a loop to calculate the IF first for MZ Twin 1 (751 participants). The problem is that this model has thresholds and I cannot add the object “threG” because it has 14 columns. I had the same problem with mean but I created and added to the model this:

mean2 <- mxAlgebra( expression= mean[,1:7], name="mean2")

I wonder if there is any way to add this:

threG2 <- mxAlgebra( expression= threG[,1:7], name="threG2" )

To the model without re-running or any way to modify the model to have this:

mxExpectationNormal("as","mean2","threG[,1:7]",dimnames=selVars[1:7])

Thank you so much in advance.

With all good wishes

Replied on Fri, 08/30/2024 - 09:32

lf-araujo Joined: 11/25/2020

Overloading?

It is difficult to help without a minimal working example. Does this work?

modelIPAE2 <- mxModel( fitIPAE,threG2, name="mulIPAE2c" ) fitIPAE2 <- mxTryHardOrdinal( modelIPAE2, intervals=F, extraTries=11 )

fitIPAE2 <- mxModel(fitIPAE2, mxExpectationNormal("as","mean2","threG[,1:7]",dimnames=selVars[1:7]))

Or with overloading:

modelIPAE2 <- mxModel( fitIPAE,threG2, name="mulIPAE2c" )


fitIPAE2   <- mxTryHardOrdinal( modelIPAE2, intervals=F, extraTries=11 )

fitIPAE2 <- fitIPAE2 + mxExpectationNormal("as","mean2","threG[,1:7]",dimnames=selVars[1:7])

Replied on Tue, 09/03/2024 - 03:13

JuanJMV Joined: 07/20/2016

Thank you so much for this

Thank you so much for this @If-araujo

That will work but if I do that I have to run the model again (more than one week).

Here you have a simple example using some simulated data. It is a univariate model an I have added a new object "mean2" but re-running the model. My question is if there is any way to add "mean2" into the model without re-running it.

Thank you so much in advance!


library(OpenMx)

library(psych); library(polycor)

source("miFunctions.R")

mxOption(NULL,"Default optimizer","CSOLNP")

set.seed(1)

cov.matone <- matrix(c(1, .7,

                       .7, 1), nrow = 2)
dfMZ <- data.frame(MASS::mvrnorm(n = 1e4,

                                 mu = c(0, 0),

                                 Sigma = cov.matone))
cov.mattwo <- matrix(c(1, .35,

                       .35, 1), nrow = 2)
dfDZ <- data.frame(MASS::mvrnorm(n = 1e4,

                                 mu = c(0, 0),

                                 Sigma = cov.mattwo))
cov(dfMZ)

cov(dfDZ)
# ----------------------------------------------------------------------------------------------------------------------

# Program: oneACEvc.R

# Author: Hermine Maes

# Date: 10 22 2018

#

# Twin Univariate ACE model to estimate causes of variation across multiple groups

# Matrix style model - Raw data - Continuous data

# -------|---------|---------|---------|---------|---------|---------|---------|---------|---------|---------|---------|

# Select Variables for Analysis

vars <- 'X' # list of variables names

nv <- 1 # number of variables

ntv <- nv*2 # number of total variables

selVars <- paste(vars,c(rep(1,nv),rep(2,nv)),sep="")

# Select Data for Analysis

mzData    <- dfMZ

dzData    <- dfDZ

# Generate Descriptive Statistics

colMeans(mzData,na.rm=TRUE)

colMeans(dzData,na.rm=TRUE)

cov(mzData,use="complete")

cov(dzData,use="complete")

# Set Starting Values

svMe <- 20 # start value for means

svPa <- .2 # start value for path coefficient

svPe <- .5 # start value for path coefficient for e

# ----------------------------------------------------------------------------------------------------------------------

# PREPARE MODEL

# Create Algebra for expected Mean Matrices

meanG <- mxMatrix( type="Full", nrow=1, ncol=ntv, free=TRUE, values=svMe, labels=labVars("mean",vars), name="meanG" )

# Create Matrices for Variance Components

covA <- mxMatrix( type="Symm", nrow=nv, ncol=nv, free=TRUE, values=svPa, label="VA11", name="VA" )

covC <- mxMatrix( type="Symm", nrow=nv, ncol=nv, free=TRUE, values=svPa, label="VC11", name="VC" )

covE <- mxMatrix( type="Symm", nrow=nv, ncol=nv, free=TRUE, values=svPa, label="VE11", name="VE" )

# Create Algebra for expected Variance/Covariance Matrices in MZ & DZ twins

covP <- mxAlgebra( expression= VA+VC+VE, name="V" )

covMZ <- mxAlgebra( expression= VA+VC, name="cMZ" )

covDZ <- mxAlgebra( expression= 0.5%x%VA+ VC, name="cDZ" )

expCovMZ <- mxAlgebra( expression= rbind( cbind(V, cMZ), cbind(t(cMZ), V)), name="expCovMZ" )

expCovDZ <- mxAlgebra( expression= rbind( cbind(V, cDZ), cbind(t(cDZ), V)), name="expCovDZ" )

# Create Data Objects for Multiple Groups

dataMZ <- mxData( observed=mzData, type="raw" )

dataDZ <- mxData( observed=dzData, type="raw" )

# Create Expectation Objects for Multiple Groups

expMZ <- mxExpectationNormal( covariance="expCovMZ", means="meanG", dimnames=selVars )

expDZ <- mxExpectationNormal( covariance="expCovDZ", means="meanG", dimnames=selVars )

funML <- mxFitFunctionML()

# Create Model Objects for Multiple Groups

pars <- list( meanG, covA, covC, covE, covP )

modelMZ <- mxModel( pars, covMZ, expCovMZ, dataMZ, expMZ, funML, name="MZ" )

modelDZ <- mxModel( pars, covDZ, expCovDZ, dataDZ, expDZ, funML, name="DZ" )

multi <- mxFitFunctionMultigroup( c("MZ","DZ") )

# Create Algebra for Unstandardized and Standardized Variance Components

rowUS <- rep('US',nv)

colUS <- rep(c('VA','VC','VE','SA','SC','SE'),each=nv)

estUS <- mxAlgebra( expression=cbind(VA,VC,VE,VA/V,VC/V,VE/V), name="US", dimnames=list(rowUS,colUS) )

# Create Confidence Interval Objects

ciACE <- mxCI( "US[1,1:3]" )

# Build Model with Confidence Intervals

modelACE <- mxModel( "oneACEvc", pars, modelMZ, modelDZ, multi, estUS, ciACE )

# ----------------------------------------------------------------------------------------------------------------------

# RUN MODEL

# Run ACE Model

fitACE <- mxRun( modelACE, intervals=T )

sumACE <- summary( fitACE )

fitACE$algebras

mean2 <- mxAlgebra( expression= meanG[,1:1], name="mean2") modelACE2 <- mxModel( fitACE,mean2, name="mulIPAE2c" ) fitACE2 <- mxRun( modelACE2, intervals=T ) sumACE2 <- summary( fitACE2 ) fitACE2 fitACE2$algebras

Replied on Wed, 09/04/2024 - 08:56

lf-araujo Joined: 11/25/2020

Gotcha

Now I think I understand what you are trying to do. Simple answer: no, you can't.

By adding the new means algebra the estimation step must be run. Now, I know of the existence of mxEval() and I am not sure if you can use it to add the new expression and ask the optimizer to only run that, given the already estimated model.

Replied on Wed, 09/04/2024 - 09:03

lf-araujo Joined: 11/25/2020

mxEval

So here the mxEval example:


     library(OpenMx)
     # Set up a 1x1 matrix

     matrixA <- mxMatrix("Full", nrow = 1, ncol = 1, values = 1, name = "A")
     # Set up an algebra

     algebraB <- mxAlgebra(A + A, name = "B")
     # Put them both in a model

     testModel <- mxModel(model="testModel3", matrixA, algebraB)
     # Even though the model has not been run, we can evaluate the algebra

     #   given the starting values in matrixA.

     mxEval(B, testModel, compute=TRUE)
     # If we just print the algebra, we can see it has not been evaluated

     testModel$B

So you may be able to do something like:
mean2 <- mxAlgebra( expression= meanG[,1:1], name="mean2") modelACE2 <- mxModel( fitACE,mean2, name="mulIPAE2c" ) # then

mxEval(mean2, modelACE2, compute=TRUE)

That said, I really never get mxEval to work properly, but this is certainly a skill issue.

Best,
L

Replied on Sun, 09/15/2024 - 04:05

JuanJMV Joined: 07/20/2016

Thank you so much If-araujo.

That works to make calculations but I need to use it in this line:

fsModel@expectation <- mxExpectationNormal("as", "mean2","threG",dimnames=selVars[1:7])

So, I think it must be computed.
I have tried to add it as a separate matrix but I got this error
Error en mxExpectationNormal("as", "mean2", mxMatrix(values = c(1, 2, :
'thresholds' argument must be a single matrix or algebra name

So, I think it must be computed and contained in the model.

I am re-running the model, so do not worry and thank you so much for your help!

With all good wishes,

News

Recent Posts

How to add object to a model without re-run

more info?

Hi Rob,

Overloading?

Thank you so much for this

Gotcha

mxEval

Thank you so much If-araujo.

News

Recent Posts