More consideration on IHR priors?

[SamuelBrand1]

Given that this is a big model to compute (as in a lot of compartments), and we want to do Bayesian analysis, then its probably a good idea to make strong priors where possible.

I think the priors for infection-hospitalisation-ratio (IHR) are too vague and should probably reflect known age structure in IHR?

https://github.com/cdcent/cfa-scenarios-model/blob/dfbd4ded9fddf2dafcd793f73f751e6067ccd9f2/mechanistic_model/mechanistic_inferer.py#L130-L131

The prior range here looks like prior mean 4.8% IHR with 2.5-97.5% range: 0.005% - 22.7% with no age variation.

For me thats too vague and doesn't help the sampler (or indeed the reasoning); I think we can tighten those priors and add age structure.

NB: Pre-apologies if I'm not getting something here!

[kokbent]

Yeah in the actual inference work we do, we usually bypassed the default mechanistic inferer to write a version of IHR with more stringent priors. Here's an example for our SMH work: https://github.com/cdcent/cfa-scenarios-model/blob/4d04c4a381d4d76a1f13e6b4dc45c4a4ec7d9ca7/exp/fifty_state_2202_2307_3strain/inferer_smh.py#L143

[SamuelBrand1]

Yeah in the actual inference work we do, we usually bypassed the default mechanistic inferer to write a version of IHR with more stringent priors. Here's an example for our SMH work:

https://github.com/cdcent/cfa-scenarios-model/blob/4d04c4a381d4d76a1f13e6b4dc45c4a4ec7d9ca7/exp/fifty_state_2202_2307_3strain/inferer_smh.py#L143

Wouldn't it be easier to make something that takes in priors and return the likelihood function rather than almost C&P things, but hard coding prior changes?