evgenyneu / covid19

A naive Stan model of confirmed COVID-19 cases that uses logistic function.

Using logistic function to model number of COVID-19 confirmed cases in Stan and Python

This is a Stan model for analysing time series of confirmed cases of Coronavirus (COVID-19). It uses logistic function, which I described in this blog post.

⚠️⚠️⚠️WARNING ⚠️⚠️⚠️

This is a simple model that was done for educational purposes only. It is almost certainly very inaccurate and can not be used to predict the spread of the disease.

How the model works

The program does the following:

• Downloads new data on confirmed COVID-19 cases from Johns Hopkins University Github's repository.

• Runs Stan model that uses logistic function.

• Creates a plot showing observed data with round orange markers and the model with blue line. The bright shaded region indicates model's uncertainty and corresponds to 95% DHPI (highest posterior density interval).

As new data become available daily, I will be running this script and adding the new plots to plots directory. This way we can see the changes in the data and the model with time.

Extrapolating the data

Attention, the following plot was made for fun and is complete nonsense because it was bade with unrealistic model assumptions, which are described below. The plot also extrapolates the data way far into the future, which is silly.

Assumptions

In the model we make the following assumptions:

• The number of cases can be modelled with logistic function.

• The virus is able to infect fixed number of people in total. However, it will be very interesting to make it a free parameter and let Stan estimate its distribution instead.

• All people who are infected with COVID-19 are reported.

Exclusion of China

The model does not include the data for China because the number of confirmed cases have levelled off, as shown below. Assuming these data are correct, this could indicate that the spread of the disease is under control. Well done, China!

Setup

In order to run the model, download and install all required software with the following steps.

1. Install python libraries

First, install Python 3.7 or newer, and then run:

pip install tarpan

If you are having issues with running the code, use pip install -Iv tarpan==0.3.9 instead.

2. Install Stan

install_cmdstan

Usage

To run the code, first, download this repository:

git clone https://github.com/evgenyneu/covid19.git
cd covid19

Finally, run the Python script:

python code/run_model.py

The script will run the model and save plots into the plots directory.

Data source

The data are taken from Johns Hopkins University Github's repository.

Learning stats

If you want to learn more about Stan and statistical programming, I would highly recommend Statistical Rethinking textbook by Richard McElreath.

The unlicense

This work is in public domain.