covidsimulation

Covid Epidemics Simulation - Individual-based dynamic model in python and cython with Simpy

"All models are wrong, but some are useful." George Box

About the model

This is a detailed individual-based simulation of the Covid-19 epidemic within a region. It is possible to model different age-groups and different sub-populations, that behave differently with regards to transmission and interventions. Different interventions, or sequences of interventions, can be simulated as well. Most parameters can be set as a random value within a distribution, allowing to model our current uncertainty. A calibration process is provided, in which parameters or random distributions can be optimized, effectively updating our belief about them. It is possible to plot curves for many output variables of the simulation, like expected official death counts, expected real death counts, intensive-care usage, reproductive number, etc.

The model mainly is intended for:

providing a reallistic uncertainty range about the development of the epidemic
studying the possible effects of interventions
providing insights about what factors can be contributing to the overall results observed.

An example of how to do a basic simulation can be found in https://towardsdatascience.com/model-the-covid-19-epidemic-in-detail-with-python-98f0d13f3a0e

Instructions

Clone this repo
Run setup.sh (It might require making it executable: chmod 777 setup.sh)
Check Example.ipynb to understand how it works
Setup your own regions and modify simulation parameters and constants
Whenever you change Cython code, don't forget to compile it with python3 setup.py build_ext --inplace or simply call setup.sh again

Troubleshooting

In case you have issues with clogged outputs or in case graphs don't show up, you can try enabling some nbextensions in your jupyter: (If you are not in a virtualenv, use --user instead of --sys-prefix)

jupyter nbextension install widgetsnbextension --sys-prefix --py
jupyter nbextension install plotlywidget --sys-prefix --py
jupyter nbextension enable widgetsnbextension --sys-prefix --py
jupyter nbextension enable plotlywidget --sys-prefix --py

Understanding the code

This is an IBDM for Covid-19. The model is based on the concept of Person. Each Personhas several state variables indicating her health, home location, age, etc, and also indicating the progression of the disease. All happening with uncertainty. Statistics about the whole population are gathered at the end of each day and stored as measurements.

Simpy

The simulation happens using Simpy's processes and resources. When env.process(do_something()) is called, do_something() is executed assynchronously (in simulation time). Within do_something(), executing runs until being blocked by yield env.timeout(duration), and resumes after duration days. PriorityResource manages capacity and a priority queue to use it.

Geosocial model

We expect a Person to have a higher chance of interaction with people who live close to them. Also, we expect most people will interact more often with other people from the same social group, and even age group. At the moment, this was achieved approximating those interactions as sampling a few random individuals, and then choosing the one who is the closest in a multidimensional space. This simplification can be replaced by a more accurate model in the future.

Cython

To achieve a good resolution, a few hundreds of thousands of Persons need to be simulated. Using pure Python would be slow and memory bound. By using Cython to define Person objects, speed was significantly improved, and it is possible to simulate 1M people with less than 1GB RAM per core.

ToDo's

Better document model setup, training and usage.
Model more regions, from different countries.
Model more possible interventions (e.g. intermitent social distancing, rotational work schedules, etc.)
Improve the probabilistic (calibration) model, modeling the information gain from different checks, and enabling transfer of believes between regions.
Model hospital transmission to health workers and to other patients.
Improve model of interaction between people:

Specifying different types of interaction between people
Creating sub-population and groups within which interactions can occur (e.g. school class)
Specifying different rates of social interaction for each type and age group ...

ArroyoAndre/covidsimulation