Initial Value Rejection with CVODE

[Arya Pourzanjani]

I'm trying to fit parameters to a 4 state ODE and I get the error below. The Stan and data files are attached. I'd like to learn the control flow of a Stan program so I can figure out things like this on my own. What's the best way to compile Stan to run with GDB, or am I stuck putting print statements in Stan source code and recompiling every time?

Arya$ coag/rar diagnose data file=coag/rar.data.R 

method = diagnose

  diagnose

    test = gradient (Default)

      gradient

        epsilon = 9.9999999999999995e-07 (Default)

        error = 9.9999999999999995e-07 (Default)

id = 0 (Default)

data

  file = coag/rar.data.R

init = 2 (Default)

random

  seed = 3933974504

output

  file = output.csv (Default)

  diagnostic_file =  (Default)

  refresh = 100 (Default)

Informational Message: The current Metropolis proposal is about to be rejected because of the following issue:

Exception thrown at line 84: CVode failed with error flag -1

If this warning occurs sporadically, such as for highly constrained variable types like covariance matrices, then the sampler is fine,

but if this warning occurs often then your model may be either severely ill-conditioned or misspecified.

[Michael Betancourt]

Stan is like any other C++ program and can be run through GDB with enough

effort, although you’ll have to be careful because of all the templates.  Print

statements are typically much faster than trying to following GDB as it bounces

between Stan and CVODE.

The -1 error states that CVODE hit the max_num_steps threshold that you

specified (See Section B.2 of the CVODE manual), which is not uncommon

out in the tails of the parameters. You can increase the threshold from 1e6 to

1e8 or the like or find a better initial condition, but your model looks sufficiently 

complex that I would step back and start with something simpler.  The bounds

are not helping anything — use soft contains with priors instead.  The 1e-11

in the ODE specification is a sure sign that you can rescale your parameters

so that they’re all O(1) which could make huge improvements. 

--
You received this message because you are subscribed to the Google Groups "Stan users mailing list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to stan-users+...@googlegroups.com.
To post to this group, send email to stan-...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
<rar.stan><rar.data.R>