v-a-s-a / bayesian_sir

Covid SIR modeling with bayesian parameter estimation.

Bayesian parameter estimates for non-linear systems of differential equations

Please note, the following work is for educational purposes only. If you'd like to draw valid inferences, please consult an infectious disease epidemiologist. Modeling the outbreak is hard, and this data must be used responsibly (thanks to @simonw for raising these points).

This was inspired by Thomas Wiecki's work on fitting PyMC3 models to the COVID data, as well as the elegant DiffEqBayes.jl package.

Data

Case data comes from Johns Hopkins Center for Systems Science and Engineering. Country population data comes from the European Centre for Disease Prevention and Control .

Model

Currently, the model is hierarchical SIR model, with infection and recovery rates grouped by country. The system is specified and solved using DifferentialEquations.jl and parameters are infered using Turing.jl.

Results

The following show distributions of disease trajectories over time. Thick lines represent the expected posterior distribution of disease trajectory.

Data/fit from 20.04.2020: