Daily incidence data generating process convergence issues

[TimothyWillard]

The model has a hard time converging even for some datasets. Unsure of what the issue is or how to test/fix it, but it poses a major blocker to usability.

[TimothyWillard]

I suspect that this has to do with daily incidence being drawn from a gamma and then approximating the observations as being drawn from a normal with a mean of the sum of those gammas with a really small variance. This likely biases the fitting (although not by much?) but adds a lot of computational complexity. Making this guess based on the appearance of these issues coinciding with the OOP model interface. Alternatives include:

1. Finding some approximation for the sum of gamma distributed variables and then manually adding that to the total likelihood when fitting, or
2. Looking for a two parameter continuous and positive distribution with a closed form sum.

Decisions here do affect the ability to configure the observational noise and add "indirect" observations down the road, so need to be conscious of that. See #22, #48.