Interface for using SMCPy without explicit model function

[bronevet-abc]

Hi, I'd like to use SMCPy in a context where the "model to evaluate" is located on one or more remote machines. As such, I need to collect a set of samples to evaluate from SMCPy, send them over for remote evaluation and then get the outcomes and feed them back to SMCPy. It looks like this use-case can be supported via the sample() method in the samplers and direct calls to the methods of SMC. Is that correct? If so, is there an example I could follow to make this work? Thank you!

[peleser]

Hey @bronevet-abc! Can you give me a little more detail on your problem? While it is technically possible to manually pass results between machines by breaking up some of the sampler functionality, I think this could be a pretty big headache. There are alternatives depending on your exact setup. For one, if you can use MPI to achieve your goal, SMCPy already supports this. See this example. If that doesn't work for you but you have some sort of automated pipeline (e.g., ssh-based) that allows you to send inputs to the machine and return results in an automated fashion, you could wrap that functionality in a model class or build a special MCMC class to handle this. If you're using a cluster with a job submission queue, you may be able to develop a model wrapper that submits jobs to the queue and waits, etc.