Can't solve dimname associated error

This is pretty easy to fix. You can just provide the dimnames in your call to `mxAlgebra()`. Suppose you were analyzing five variables, with the following five column names: "x1","x2", "x3", "x4", and "x5". You could then do something like,

mxAlgebra(cov_matrix %*% t(variances) %*% cov_matrix, name = "predicted_cov", dimnames=c("x1","x2","x3","x4","x5")

. I notice that you don't have a header row in your dataset. You're going to have to create one, and use `read.csv()` with argument `header=TRUE`, so that OpenMx knows which variables it's going to be analyzing.

The below version of the model "works" in the sense that there are no errors. However, the model-implied covariance matrix at the starting values is not a valid covariance matrix.


pheno2 <- read.csv("pheno2.csv")
library(OpenMx)

colnames(pheno2) <- gsub("\\.", "_", colnames(pheno2))
any(is.na(pheno2))

matrix_data <- matrix(as.double(unlist(pheno2)), ncol = 7)

# Set row names for each observation
rownames(matrix_data) <- 1:nrow(matrix_data)

# Set column names for each variable
colnames(matrix_data) <- colnames(pheno2)

# Now, create the OpenMx data object
mxDataObject <- mxData(observed = matrix_data, type = "raw")

num_variables <- 7
cov_matrix <- mxMatrix(type = "Full", nrow = num_variables, ncol = num_variables, free = TRUE, values = 1, name = "cov_matrix")
means <- mxMatrix(type = "Full", nrow = 1, ncol = num_variables, free = TRUE, values = 0, name = "means")
variances <- mxMatrix(type = "Diag", nrow = num_variables, ncol = num_variables, free = TRUE, values = 1, name = "variances")

expectation <- mxExpectationNormal(covariance = "predicted_cov", means = "means", dimnames=c("NS","RA","RD","P","SD","C","ST"))

model <- mxModel("MyModel",
cov_matrix,
means,
variances,
mxAlgebra(cov_matrix %*% t(variances) %*% cov_matrix, name = "predicted_cov"),
expectation,
mxDataObject,
mxFitFunctionML())
fit <- mxRun(model)



Running MyModel with 63 parameters
Error: The job for model 'MyModel' exited abnormally with the error message: fit is not finite (The continuous part of the model implied covariance (loc2) is not positive definite in data 'MyModel.data' row 1. Detail:
covariance = matrix(c( # 7x7
7, 7, 7, 7, 7, 7, 7
, 7, 7, 7, 7, 7, 7, 7
, 7, 7, 7, 7, 7, 7, 7
, 7, 7, 7, 7, 7, 7, 7
, 7, 7, 7, 7, 7, 7, 7
, 7, 7, 7, 7, 7, 7, 7
, 7, 7, 7, 7, 7, 7, 7), byrow=TRUE, nrow=7, ncol=7)
)
In addition: Warning message:
In model 'MyModel' Optimizer returned a non-zero status code 10. Starting values are not feasible. Consider mxTryHard()


The first question I have for you is: what do you mean by "I'm trying to create a covariance matrix with my data?"

Second, what are you hoping this algebra will do?


mxAlgebra(cov_matrix %*% t(variances) %*% cov_matrix, name = "predicted_cov")


So, is your question solved for now?

The problem with my previous script was just that the starting values implied an improper covariance matrix: a 7x7 matrix of all 7s.

Replacing this line

cov_matrix <- mxMatrix(type = "Full", nrow = num_variables, ncol = num_variables, free = TRUE, values = 1, name = "cov_matrix")

with these lines

set.seed(41)
rvals <- matrix(runif(num_variables*num_variables), num_variables, num_variables)
rvals <- rvals %*% t(rvals)
cov_matrix <- mxMatrix(type = "Full", nrow = num_variables, ncol = num_variables, free = TRUE, values = rvals, name = "cov_matrix")

fixes the starting values issue.

Are there any other parts I should change?

I really have no idea if there are parts you *should* change. It depends to a large extent on how the model you are specifying relates to the model you actually want to fit. I still don't know what model you'd like to fit and how this script relates to it.

This warning is generally saying "The optimum solution doesn't really look right (non-convex Hessian); consequently, don't trust your standard errors; your model might not be identified.". In this case, my guess is the model you specified has little to do with the model you want to specify and little to do with any reasonable model for these data. We can't help you with that unless you give us more information.

For the purposes of clarity, we have solved your initial question: you no longer have a problem with dimnames.

If you have further questions relating to a model you'd like to fit, we will all be happy to help you with them. Please start a new thread with whatever that question is. If you do so, please be as specific as you can about the model you are trying to fit and what questions you have.