Error message when using get() to return values of OpenMx object

Posted on
No user picture. Veronica_echo Joined: 02/23/2018
Hi everyone,

I am trying to return values of an OpenMx object from optimized mxModel by get() (more details could be found in attached). Is there any other way to return the value of an OpenMx object by string? Thanks in advance.

Replied on Sun, 11/18/2018 - 22:34
Picture of user. mhunter Joined: 07/31/2009

I'm not sure what you mean by "return the values of an OpenMx object". You can save the final state of a model to a file with `mxSave` or get the free parameters with `coef` (see `?omxGetParameters`). You can evaluate algebras with `mxEval` and write those values to files with e.g. `write.csv`. Hopefully this helps. If not, more details please.
Replied on Mon, 11/26/2018 - 11:12
No user picture. Veronica_echo Joined: 02/23/2018

In reply to by mhunter

Hi Michael,

Thanks for your informative reply. I want to evaluate algebras; for example, I have a series of algebras named "c1mean", "c2mean", ..., "ckmean". What I want is to create strings by paste0("c", 1:k, "mean"), change them to object names, and to evaluate them. Do you have suggestions about it? Any advice would be appreciated. Thank you!

Replied on Mon, 11/26/2018 - 12:14
Picture of user. AdminRobK Joined: 01/24/2014

In reply to by Veronica_echo

Perhaps adhere the 1 thru _k_ algebras into a single algebra with `cbind()` and/or `rbind()`, and use `mxEval()` to evaluate that single algebra?

At any rate, the function `mxEvalByName()` may be useful to you, in case you didn't know about it already.

Replied on Mon, 12/17/2018 - 23:40
No user picture. Veronica_echo Joined: 02/23/2018

In reply to by AdminRobK

Dr. Kirk,

After having point estimates by mxEvalByName(), I still want to have the SEs, which I thought can be realized by mxSE(). Yet I got some errors as attached, which might be due to mxAlgebra(). I tried to find the string version of mxSE(), but failed. How can I fix this error? Any advice would be appreciated.

Replied on Mon, 11/26/2018 - 12:56
Picture of user. tbates Joined: 07/31/2009

you probably don't want to use get(). things inside models are not in the global name space (same reason get(mpg) won't return the mpg column of mtcars dataframe...

You might want mxEval() and mxEvalByName()

So, if we bang out a quick twin model, and interogate it:


require(umx)
data(twinData) # ?twinData from Australian twins.
# Pick the variables
selDVs = c("ht")
mzData <- twinData[twinData$zygosity %in% "MZFF", ]
dzData <- twinData[twinData$zygosity %in% "DZFF", ]
m1 = umxACE(selDVs = selDVs, sep = "", dzData = dzData, mzData = mzData) # -2ll= 9659, a1 = .92

#What's A?
mxEval(top.A, model = m1)
[,1]
[1,] 0.003727328
# or
mxEval(A, model = m1$top)

# or
mxEvalByName("top.A", model = m1)

Replied on Mon, 11/26/2018 - 14:18
Picture of user. mhunter Joined: 07/31/2009

I think Veronica wants mxAlgebraFromString() and then mxEval() that algebra. It it something like this.


astring <- paste0("cbind(", paste0("c", 1:k, "mean", collapse=', '), ")")
# e.g. for k=7 astring is
# "cbind(c1mean, c2mean, c3mean, c4mean, c5mean, c6mean, c7mean)"
palg <- mxAlgebraFromString(astring, name='Jerry')
amodel <- mxModel(... stuff ..., palg)
# ... put palg in your model
mxEval(Jerry, amodel)

Of course, there is another consideration. You may want to re-think your model construction so that you don't need to do this. For example, you could have a matrix like this


amat <- mxMatrix('Full', nrow=k, ncol=1, free=TRUE, labels=paste0('c', 1:k, 'mean'), name='JerryM)

Anywhere you want to use 'c3mean' just refer to it by name. This way you don't need to construct 'Jerry' because you started with 'JerryM'.

Replied on Thu, 11/29/2018 - 17:20
No user picture. Veronica_echo Joined: 02/23/2018

Thanks for all kind and detailed replies! I created several expressions (c1mean, c2mean, ..., ckmean) using mxAlgebraFromString() and want to return their values by character name. I tried mxEvalByName() and it worked well. I really appreciate it.