Several core users report models (typically running for many hours) returning as if the run was successful, but the returned model is marked as never run, contains no output.
I think all cases involve ordinal/binary data or joint ordinal.
CSOLNP is known to have this behavior, May affect NPSOL as well: need to check and update this bug report
Happens with 1 core as well as several.
Mike suspects optimiser finds an -Inf log likelihood and quit.
We should at least return a code red and the parameters from the lowest -2ll (constraint-satisfying) run rather than confuse the user by claiming they never tried to run the model ( which is what summary says
Better yet of course to be able to run realistic joint ordinal models without error...
Comments
#1
Does anyone have a script that produces this behavior? I don't doubt that it exists (existed). There's just nothing I can do without more concrete evidence.
#2
So long ago that no one can verify this issue's existence. Maybe fixed, maybe not. We'll keep an eye out for it, but close it in the meantime.