Dear Stan Team,

I am working on a larger Dataset of different Time series of experiments. Larger number of Observables the maximum calculation time on my of 8days cluster runs out. I would like to profile whether my assumptions about the time spend calculation the likelihood is the bottleneck of my stan model (since there are many data points and each of it requiring some matrix inversion, multiplications etc…). It is mathematical obvious that one could parallelize those for loops over each time trace to different CPUs while calculating each chain on a different node.

Are there any Helloworlds or Tutorials for those issues? I haven t found a helping documentation for manipulating the c++ code of stan model. The pystan stanc function delivers me some c++ code for some class but I don t really find a starting point to understand this class.

Especially where do I find some main function in which I can retrace the Programmblocks like the model{}and the transformed Parameterblock{}?

Where do I find of the C++ code which distributes the chains to different cores which I would like to change to distribute over nodes of the cluster?

Are there ways yet, to calculate the trajectory on the parameter space in parallel for a model with Hidden State space properties… such as Hidden Markov or a Kalmann filter?

Thanks for your comments and hints in advance.

Jan