Sirum
Repository to experiment with SIR-like models. The purpose
is not to develop new SIR-models but rather to create a thin wrapper around
existing SIR-implementations to allow for easy experimentation.
To start with:
- We take the minimal SIR-model and combine it with a blackbox-optimiser like Optuna or Hyperopt
- Expand the complexity of the SIR-model (SEIR, SEIRSPLUS, SIRF, SIS, SIRD, etc.)
- Try different model paradigms?
Research questions:
- what is the required immunity
- how to include/extract the required IC-capacity in/from the model?
I.e. we have the following basic components: {{intial conditions}-> {SIR-model}} <-> {Optimisation/fitting}
Problems:
- the population is either assumed perfectly mixed or the parameters are identical over all stratifications -> this will lead to a gross over-estimation of the number of death/critically ill if the mildly affected patients are separated from the weak patients: SEIRSplus handles this.
- it does not have spatially separated population hubs that interact
- it does not have spatially centrally connected points where people interact
Suggestions for improvement:
- the mild, moderate and clinical patients should be physically separated in 3 models that interact
- those 3 models should operate in individual population hubs
- those population hubs should have central connection points
online calculators and dashboards:
Crowd initiatives:
Kaggle competitions: https://www.kaggle.com/covid19
Good notebooks:
Literature sources:
Data sources:
Specific papers:
Explanations: