Main menu
You are here
omxMnor() problem
Mon, 02/25/2013 - 22:54 — neale
Funky output with non-positive definite input matrix. May be funky otherwise too. See:
but note that there are some issues with the script, which I have fixed in the attached. Data here also, for convenience.
https://openmx.ssri.psu.edu/thread/1949#comment-4591
| Attachment | Size |
|---|---|
 curiousProblem.R | 2.51 KB |
 dz.csv | 1.58 KB |
Reporter:
Created:
Mon, 02/25/2013 - 22:54
Updated:
Wed, 12/02/2015 - 16:18
- Log in or register to post comments
Printer-friendly version
Comments
#1
It is unclear based on this post what the real problem is and what the desired behavior is. I've attached a script that demonstrates the odd behavior. (The old one no longer did due to data sorting and which row is returned as the result of an algebra after running.) It still depends on the data file Neale attached.
Remark: throwing an error in the backend when a correlation greater than 1 is observed causes errors in the test suite.
Question 1: Do we just want a frontend error for non-positive definite cov matrices?
Question 2: Is the problem that optimization led to an expected non-positive definite cov matrix?
#2
See https://github.com/OpenMx/OpenMx/commit/2688ee61cda7381ddadbde84cb7e27ef026b7f16 for the fix.