Prior predictive checks: models all comparable?

[SamuelBrand1]

No description provided.

[SamuelBrand1]

My intention here is that there is a sub-directory of test that has prior predictive check visualisations. Let me know any extra analysis we want here?

[seabbs]

That sounds very sensible to me

[SamuelBrand1]

This might just be a placeholder... locally I render using Literate... so prior pred checks could end up in the docs

[seabbs]

ah yes that would also be nice

[seabbs]

or as a doctest?

[SamuelBrand1]

or as a doctest?

We can use Documenter.@example in the doc strings to do this.

[seabbs]

Is this 1 done. 2. part of the pipeline or 3. needing a rewrite for where we are now?

[SamuelBrand1]

Its 3. needing a bit of a rewrite.

In particular, as discussed f2f we need to choose appropriate priors dependent on the infection generating process.

For example, I think its reasonable to say in any given scenario that the exponential growth rate is with close to probability one in $[-0.35, 0.35]$ (e.g. doubling time is less than 2 time steps) whereas log(Infections) would be on a somewhat different scale.

This needs to be written in and prior predictive checking done to validate that our prior beliefs about the kind of infection processes we are modelling is accurately reflected in our parameter priors.

[SamuelBrand1]

Because we are considering different mean generation times we should probably consider our priors in light of that too.

[SamuelBrand1]

It turns out that switching to daily/one time step increments seems to create overflow edge cases (e.g. #433 ). This highlights the need to complete this issue.

It is (obviously) also part of the analysis plan.