This repo contains RL environments that model epidemic control problems. Additionally in examples_sir
, there are models and scripts to recreate the results in "One-Shot Epidemic Control with Soft Actor Critic".
We have the following environments:
sir-v0
is based on the Morris et al. paper. There are different modes of intervention (specifically, full suppression and fixed control), which have different action spaces; but generally, the agent inputs a real-valued vector that specifies the non-pharmaceutical intervention -- e.g. when to intervene and how severely the agent should reduce transmission. The observation space is a real-valued vector that reports S, I and R0.sir_multi-v0
is also based on the Morris et al. paper. But here, instead of allowing for one-shot intervention as insir-v0
, the agent has the ability to intervene on a weekly basis. The agent can decide how severely to reduce transmission. The agent has an intervention budget of 8 weeks.
The environments can be installed and run from the command line via:
pip install gym_epidemic
If you are cloning this repo from github; from the current directory run:
pip install -e .