This repository implements an agent-based pandemic simulator to analyse and optimize mitigation policies that manage economic impact of pandemics without overwhelming hospital capacity. Unlike traditional agent-based models, the simulator is able to model fine-grained interactions among people at specific locations in a community, at a level of detail that allows the spread of the disease to be an emergent property of people’s behaviors and government’s policies. An interface with OpenAI Gym environment is also provided to support an RL-based methodology for optimizing mitigation policies.
The simulator is developed in collaboration between AI researchers and epidemiologists and is designed to study pandemics in general. However, the current focus is to support the struggle against dealing with the ongoing COVID-19 pandemic.
At a high level, the simulator currently models:
- a population with a realistic age distribution, where each person has a distinct pre-defined programmable stochastic (or deterministic) behavior routine
- several heterogeneous locations like schools, homes, stores, bars, restaurants, etc., with different characteristics, such as open times, capacity, contact rates, etc.
- testing with false positive/negative rates
- imperfect public adherence to social distancing measures
- contact tracing
- variable spread rates among infected individuals
If you plan to use the simulator in your research, please cite us using:
@misc{kompella2020reinforcement,
title={Reinforcement Learning for Optimization of COVID-19 Mitigation policies},
author={Varun Kompella* and Roberto Capobianco* and Stacy Jong and Jonathan Browne and Spencer Fox and Lauren Meyers and Peter Wurman and Peter Stone},
year={2020},
eprint={2010.10560},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
Clone the git repo, install pip (on Ubuntu, you can just do apt-get install python3-pip) and run pip install.
See setup.py for a list of python package requirements that will automatically be installed.
foo@Foo:~$ git clone https://github.com/SonyAI/PandemicSimulator.git
foo@Foo:~$ cd PandemicSimulator
foo@Foo:~/PandemicSimulator$ python3 -m pip install -e . from tqdm import trange
import pandemic_simulator as ps
# the first thing to do at the start of any experiment is to initialize a few global parameters
ps.init_globals() # you can pass parameters, like a random seed, that are shared across the entire repo.
# init locations
home = ps.env.Home()
work = ps.env.Office()
# init a worker
person = ps.env.Worker(
person_id=ps.env.PersonID('worker', age=35), # person_id is a unique id for this person
home=home.id, # specify the home_id that person is assigned to
work=work.id, # specify the id of the person's workplace
)
# init simulator
sim = ps.env.PandemicSim(locations=[work, home], persons=[person])
# iterate through steps in the simulator, where each step advances an hour
for _ in trange(24, desc='Simulating hour'):
sim.step()See scripts/tutorials/ for in-depth tutorials on how to use the simulator.
/docs -- API documentation plus hierarchy. Build it locally by calling make html inside docs folder. Requires
sphinx autodoc package. Installation instructions: https://www.sphinx-doc.org/en/master/usage/installation.html
/python/pandemic_simulator/ -- Source files
/scripts -- Tutorials, calibration-scripts and evaluation scripts
/test -- unittests
If you have any questions regarding the code, please feel free to contact Roberto Capobianco (roberto.capobianco@sony.com) or Varun Kompella (varun.kompella@sony.com).