Results Policy Comparison

R-based policy

Run:

python main.py --num_population 10000 --hospital_capacity_proportion 0.1  --imported_cases_per_step_range 0 --powers 1.0 --action_frequency 1 --tags no_clipping --horizon 96 --distr_family poisson --policy-comparison --no-policy-optimization --dynamics SIR

Results:

>>> list(zip(rs, values))
[(0.5, -15372.0), (0.8, -8178.0), (0.9, -6848.66666666668), (1.0, -6107.0), (1.1, -7741.303030303045), (1.2, -10137.636363636366), (1.5, -inf), (2.0, -inf), (2.5, -inf)]

Of all fixed R values, R=1 gives best overall reward.

Target Case-based Policy

Starting with 10 cases:

>>> target_cases = [5, 10, 100, 200, 500]
>>> list(zip(target_cases, values))
[(5, -2264.571428571429), (10, -6521.23376623377), (100, -8739.323232323237), (200, -8130.69696969697), (500, -7986.484848484849)]

Having 5 cases per day does the best.

Different Starting Number Infected

Starting with 1,000 infections / day (out of 10,000 population).

python main.py --num_population 10000 --hospital_capacity_proportion 0.2 --initial_fraction_infected 0.1  --imported_cases_per_step_range 0 --powers 1.0 --action_frequency 1 --tags no_clipping --horizon 96 --distr_family poisson --policy-comparison --no-policy-optimization --dynamics SIR

Result:

>>> list(zip(possible_case_levels, values))
[(5, -7337.673881673889), (10, -7748.057720057726), (100, -8980.212121212127), (200, -8326.727272727274), (500, -8092.0)]

>>> possible_case_levels = [5, 10, 20, 50, 100, 200, 500, 1000]
>>> dict(zip(possible_case_levels, values)
... )
{5: -3819.282828282829, 10: -8060.894660894669, 20: -7872.282828282836, 50: -8985.757575757583, 100: -9249.454545454548, 200: -8082.666666666668, 500: -8280.666666666668, 1000: -8599.151515151516}

Getting down to 500, 200, or 100 cases gets progressively worse... But then getting down to 10 or 5 cases gets progressively better.

Stochasticity is messing with the results a bit. Should make it deterministic.

Main Comparison

Elimination, under-control, or do-nothing:

{'cases=0': -3982.0, 'cases=10': -8023.670995671004, 'cases=100': -9115.45454545455, 'cases=1000': -8486.151515151516, 'do nothing': -9995.0}