Currently the SIRS simulation study uses noisy measurements of the daily prevalence of infection. However, usually we measure incidence. To make this example closer to the real-world use-case, it would be nice to use noisy estimates of daily incidence rather than prevalence.
Changes should entirely be limited to the observation model, nothing else needs change. It is in-scope to fix up the simulation, inference and visualisation of the results. Generating any new visualisations or results is definitely out-of-scope. Importantly, keeping backwards compatibility with old prevalence results is out-of-scope.
Suggested solution
[ ] Update the [[file:./SIRS_Results/particle-filter-example-sirs/readme.org][readme]] to describe the observation model. This should still be binomially distributed but with the total incidence as the expected value instead of the prevalence. Currently the manuscript says "truncated binomial", but looking at the code, I think this is meant to be "negative binomial", I'll need to investigate further.
[ ] Update the [[file:./SIRS_Results/partile-filter-example-sirs/run-simulation.py][simulation code]] to work with incidence rather than prevalence.
[ ] The [[file:./SIRS_Results/particle-filter-example-sirs/run-inference.py][inference code]] will also need to be updated since this has some prevalence specific details. This calls a plotter function so unclear if that will still work.
Currently the SIRS simulation study uses noisy measurements of the daily prevalence of infection. However, usually we measure incidence. To make this example closer to the real-world use-case, it would be nice to use noisy estimates of daily incidence rather than prevalence.
Changes should entirely be limited to the observation model, nothing else needs change. It is in-scope to fix up the simulation, inference and visualisation of the results. Generating any new visualisations or results is definitely out-of-scope. Importantly, keeping backwards compatibility with old prevalence results is out-of-scope.
Suggested solution
[ ] Update the [[file:./SIRS_Results/particle-filter-example-sirs/readme.org][readme]] to describe the observation model. This should still be binomially distributed but with the total incidence as the expected value instead of the prevalence. Currently the manuscript says "truncated binomial", but looking at the code, I think this is meant to be "negative binomial", I'll need to investigate further.
[ ] Update the [[file:./SIRS_Results/partile-filter-example-sirs/run-simulation.py][simulation code]] to work with incidence rather than prevalence.
[ ] The [[file:./SIRS_Results/particle-filter-example-sirs/run-inference.py][inference code]] will also need to be updated since this has some prevalence specific details. This calls a plotter function so unclear if that will still work.
[ ] To track incidence will require the [[https://pypfilt.readthedocs.io/en/latest/settings/index.html#particle-filter-settings][particle history matrix]] but this also seems a bit confusing. Since we allow for (R) to transition back to (S) so we need to add a new compartment to track cumulative infections and then take the difference in this.