You are here

Home

tell people what is actually happening

Tue, 11/11/2014 - 17:27 — tbates

re user issue https://openmx.ssri.psu.edu/thread/1111

As Mikee N says,

if(fit function < definite & evaluations <3 & free parameters > 0){
   message("your starting values are too far off reality for me find my way home...
You might try setting the means to their actual values (that often flumoxes me), and perhaps get the covariances closer to realistic values for me, but certainly start by setting each variable's mean to its actual value to give me a hint..." 
}

re mxTryHard()

Would it be wise to set scale to .05 by default?
Should mxTryHard (optionally?) set means to the phenotypic value?

Reporter: 
tbates
Created: 
Tue, 11/11/2014 - 17:27
Updated: 
Wed, 11/12/2014 - 17:25

Comments

Tue, 11/11/2014 - 20:03 — AdminJosh

Oh, oh, I think I get it. Try SVN 3999

Wed, 11/12/2014 - 11:24 — tbates

That's much better! Working for me on a linux build from trunk.

We still print the same error, but now give a supplementary warning in addition.
I wonder if this might be better:

Running Univariate Regression of y on x1
Error: The job for model 'Univariate Regression of y on x1' exited abnormally:
Optimizer returned status code 10. Starting values are not feasible. Consider mxTryHard()

instead of this?

> mxRun(uniRegModel)
Running Univariate Regression of y on x1
Error: The job for model 'Univariate Regression of y on x1' exited abnormally
with the error message: MxComputeGradientDescent: fitfunction Univariate
Regression of y on x1.fitfunction is not finite (Expected covariance matrix is 
non-positive-definite)
 
In addition: Warning message:
In model 'Univariate Regression of y on x1' Optimizer returned a non-zero status
code 10. Starting values are not feasible. Consider mxTryHard()
Wed, 11/12/2014 - 11:41 — AdminJosh

Yeah, I thought about that. That's a much more invasive change though. We currently report all optimizer status codes as a warning. I think it would take a good deal more justification to special case the reporting of this particular status code.

Wed, 11/12/2014 - 17:25 — AdminRobK

> re mxTryHard()

> Would it be wise to set scale to .05 by default?
Why, exactly? That seems too small to me. Maybe I'm missing something.

> Should mxTryHard (optionally?) set means to the phenotypic value?
That wouldn't really be practical in general--like if the mean is conditioned on definition variables, or if the expected mean is an algebra.

Copyright © 2007-2024 The OpenMx Project