[VoxBo] bad regress jobs

[Daniel Y Kimberg]

Rachel Diana wrote:
> I have a GLM in which all of the regress jobs have gone bad.  I tried
> re-running the GLM without using my 1/f noise model (created using
> vbfit) and then the jobs ran successfully.
> I have attached both my 1/f file and the log file for one of the bad
> regress jobs.
> Has anyone seen this before?

The very descriptive "error 204" means there was an issue with the
traces file, typically the file will have three "nan" lines.  Usually
this can be traced back to linear dependence among the variables in
the G matrix, which shouldn't differ between the two GLMs.  Can you
double-check that the G matrices are identical?  Actually, if you load
them up in gdw, you can check for linear dependence using one of the
menu options.  I can't think of how this could be caused just by
including the 1/f, but that doesn't mean it isn't possible.  If the G
matrices are really the same, I'd probably have to see the whole glm
directory to figure out what's going on.

dan