To run model fitting, there are two scripts available.
To improve model stability, the python script cooptimize_models.py should first be run to optimize shared parameters across multiple geographical regions. Currently, the only shared parameter across regions is the number of days it takes for an infected individual to recover.
Inside the config folder are a number of sample json files, which specify
the several different groups of locations on which to run model fitting. The suggested
config file to use for co-optimization is late_stage_areas as it contains countries and US locations
in later stages of the curve.
python cooptimize_models.py --specfile late_stage_areas --processes 4
Because the code requires grid search followed by independent local optimization runs,
it is beneficial to utilize multiple processors/cores. Add option --processors N
Once the optimal shared parameter is obtained, you can use it as an argument into the script below (--recovery_days).
To generate a model for a set of regions, create or reuse a json file like the ones in the config directory, and place it in the config directory. Then run the fit_model.py script, e.g.:
python fit_model.py --specfile metro_areas --recovery_days 20 --processes 4
This should generate a csv in your data folder with the title corresponding to your
--specfile parameter, e.g. metro_areas.csv.
This file can then be used with any of the python notebooks (.ipynb extensions).
Please see fit_model file for other options. For example, you can set batch_first_end_date and batch_last_end_date to do incremental writes to the csv by date.