Functionalise stan code

[seabbs]

The stan code is represented partially in functions and partially within the main model. Several sections (like the mapping from infections -> reports, the GP, update Rt etc) could be easily rewritten into stan functions. This would help support future model developments as adding extensions/reusing code will be easier.

Examples of future extensions this will support are:

• Fitting to multiple data sets at once.
• Adding layered gaussian process (both adding various different kernels and adding controlling GP hyper parameters additional GPs).
• Adding complexity the Rt generation via additional coefficients and scalings.

[medewitt]

I'm trying to think this through, so this might be half baked, but worth mentioning. In compiling the cmdstanr version you can reference function directories. In theory we could then expose any of the nice functions to users to write their own primary Stan model while pulling functions from EpiNow2...

estimate_infections2.stan (something user made)

functions{
#functions/convolve.stan
}
model{
user written model
}

Then when the user specified the model in the epinow or whatever piece of the API, internally on compile in CmdStanR, they will have those functions.

model <- cmdstanr::cmdstan_model(file.path(app_loc, "estimate_infections2.stan"),
                                     include_paths = file.path(app_loc))

[seabbs]

That sounds extremely useful. As we move forward the model will hopefully become increasingly modular so this be potentially a real benefit.