Toy model generate data #33
Conversation
…erating rather than making inference on data
…the epimodel struct
…dified functions and StatsPlot test dep
There was a problem hiding this comment.
I've just done a sight read and this generally looks good. Overall there is more model structure in log_infections than I was expecting (I thought it would just be a version of EpiModel with a defined map from the latent process to infections but it looks like it has more custom code than that.
Will check out and run in a bit and circle back
This is actually not a bad suggestion. If the |
|
In a f2f conversation we decided to proceed with this as is and to then open additional issues to improve documentation (#35) and to make the model implementation better show case Julia features and to be more DRY. |
As per Issue #14 and issue #15 it is necessary to have at least one fully defined model.
In this case, each model has two main parts:
Turingmodel defining how an underlying infection time series is generated from a latent process, and how the underling infection time series defines the log-likelihood of any observed case data, e.g. after a delay.Turingmodel defining the latent process, which interacts with the main model via@submodel.The PR covers defining 1. directly generating infections by
exptransforming the latent process, and 2. that latent process is a random walk with random step size variance and random initial condition. This also acts to test the underlying logic of using submodels to create modularity.Unit tests are here and here.
An example script that runs in
TestDevmode is here.