# Error: [...] fit is not finite

4 posts / 0 new
Offline
Joined: 02/13/2020 - 20:55
Error: [...] fit is not finite
AttachmentSize
12.04 KB
11.93 KB

Dear OpenMx experts/users,

I am using an ACE univariate script with AE, CE and E nested models. I think that the script itself is good and makes sense, however, using a relatively small dataset (40 MZ pairs; 25 DZ pairs) (this is fMRI data), I get the following error message:

> # Run CE model
> CEmodel <- mxModel( ACEFit, name="CEmodel" )
> CEmodel <- omxSetParameters( CEmodel, labels="A11", free=F, values=0 )
> CEfit <- mxRun( CEmodel, intervals=T )
Running CEmodel with 7 parameters
Error: The job for model 'CEmodel' 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 'DZ.data' row 11. Detail:
covariance = matrix(c( # 2x2
-0.333084, -0.829576
, -0.829576, -0.333084), byrow=TRUE, nrow=2, ncol=2)
)
In model 'CEmodel' Optimizer returned a non-zero status code 10. Starting values are not feasible. Consider mxTryHard()

> mxTryHard(CEmodel)
All fit attempts resulted in errors - check starting values or model specification

I also tried to change the starting values but it seems not to fix the issue. Looking at the OpenMx forum, I found somebody who had the same issue ( https://openmx.ssri.psu.edu/node/4398 ), although the answer doesn't help much in my situation.

Would someone be able to indicate where the problem could come from and how to solve it?

I attached to this message the ACE script in R and txt formats.

Any help would be immensely appreciated!

Many thanks,
Arthur

Offline
Joined: 01/24/2014 - 12:15
negative variance

In your ACE model, it looks as though either 'C11' or 'E11' (probably the latter) is being estimated as negative. Then, when you fix 'A11' to zero, the model-expected variance is also negative, and is too negative to be perturbed back above zero by mxTryHard(). First, put a lower bound on 'E11', like this:

compE     <- mxMatrix( type="Lower", nrow=1, ncol=1, free=TRUE, lbound=1e-4,
labels="E11", values=5, name="VE" )

Then, instead of this (line 257),

CEmodel <- mxModel( ACEFit, name="CEmodel" )

CEmodel <- mxModel( ModelACE, name="CEmodel" )

.

Offline
Joined: 02/13/2020 - 20:55
Error: [...] fit is not finite

Dear Robert,

Thank you so much for your reply. That seems to be the problem indeed, and your solution works perfectly!

Thanks again,
Arthur

Offline
Joined: 01/24/2014 - 12:15
Good to hear.

Good to hear.