CodeForPhilly / chime

COVID-19 Hospital Impact Model for Epidemics
https://codeforphilly.github.io/chime/
MIT License
205 stars 151 forks source link

["model"] n_days is too low for fitting current_hospitalized against the curve #570

Closed jlubken closed 4 years ago

jlubken commented 4 years ago

Summary

Low n_days (e.g. 30) s too low for fitting current_hospitalized against the curve.

Additional details

Instead of running n_days + i_day (because i_day is unknown), prefit simulation only runs for n_days. In many cases this value can be too low for the rising curve to actually reach the current_hospitalized value. When this happens, the fit estimating i_day is bad.

Postfit runs i_day + n_days, but this is no good when i_day has been fitted incorrectly.

Suggested fix

As a quick fix, run prefit an additional 300 days.

A better fix is to run both prefit and postfit simulations until infected drops below 0.5. Turn n_days into a parameter that only effects the UI window, not the simulation.