Unknown error

[tommyg]

Hi,

With a CJS model I am getting the error result :

> mod.10 <- mark(koot.proc, koot.ddl, model.parameters=list(Phi=Phi.a3_ttt, p=p.a2_tt), model.name="model 10")
Error in if (x4 > x2) { : argument is of length zero

********Following model failed to run : model 9 **********

Any idea what this could be? The model assumes assumes 3 age-classes for year-specific survival and two age-classes for year specific capture probabilities. A model with 3 age-classes for year-specific survival, but only year specific capture probabilities (no age-classes) executes fine, which makes me think it is something structural. A few days ago, however, this model ran fine, but since then I've filtered a little more of the data.

I know it's hard to tackle what's breaking down without the actual data, but are there any ideas ?

Thanks.

Tommy

[jlaake]

While this is a untrapped and generic error, it is explained in the Errors section of the documentaton (C23) and this is the first error that is described. The following is an excerpt from the documentation.

The following error message or one like it that occurs when mark.exe is running
or afterwards, occurs when something is amiss with the data, model setup for
MARK or you interrupted the job:
Error in if (x4 > x2) { : argument is of length zero
********Following model failed to run : [name of model]**************

Solution: Look at the most current input and output files in the directory
to see if you can discern what happened. Error messages and the output
will often move across the screen too quickly to read but you can always
look at them with a text editor to discover the reason for the problem.
The obtuse error above occurs because the output file is incomplete and
the function extract.mark.output cannot find relevant fields in the output
file.

[tommyg]

Thanks, Jeff. The model isn't converging:


INPUT --- proc stop;

ERROR -- Numerical convergence never reached.


So there's something with the data and/or model specification on my part.

Thanks.

[tommyg]

For those interested, the work around ended up being specifying initial values with the 'initial' argument in the function mark (section c10). Similar models were fit with the same complexity in survival, but not in capture probability and vice versa. The estimates from those parameters were used as initial starting values and a few parameters sitting on boundaries were additionally specified.

Thanks to Jeff for this suggestion.