Probabalistic predictions of COVID-19, updated daily
This project is maintained by mrtommyb
This shows the number of confirmed infections reported by the Johns Hopkins University Center for Systems Science and Engineering (JHU CSSE) for a few countries I thought interesting. They have some great visualizations, you should check it out. Data comes from their github repo and is updated every day.
I took the observed number of confirmed COVID-19 cases created a model that uses logistic function to predict future cases. This is a sigmoid shape and so allows for exponential growth that slowly flattens out. The data predictions come from a Markov chain Monte Carlo analysis with priors that prevent the number of infections from being larger than the total population of the country. Also, it probably shouldn’t be taken seriously by anyone.
The figures below are auto-generated using data on the latest number of confirmed cases and extrapolate from the observations by 50 days. The peak infection rate is the last day where the number of daily cases increases and the total number of infections is over the entire epidemic. The uncertainty regions are 80% confidence intervals.
As of Jun 16, 2020, the model predicts that China infection rate peaks on Feb 07, 2020 and a total of 84.42 [84.38 - 84.47] thousand people will be infected.
As of Jun 16, 2020, the model predicts that Italy infection rate peaks on Mar 28, 2020 and a total of 237.33 [237.29 - 237.38] thousand people will be infected.
As of Jun 16, 2020, the model predicts that South Korea infection peaks on Mar 04, 2020 and a total of 12.19 [12.16 - 12.23] thousand people will be infected.
As of Jun 16, 2020, the model predicts that Iran infection peaks on Apr 13, 2020 and a total of 189.92 [189.88 - 189.97] thousand people will be infected.
As of Jun 16, 2020, the model predicts that France infection peaks on Apr 05, 2020 and a total of 194.35 [194.31 - 194.42] thousand people will be infected.
As of Jun 16, 2020, the model predicts that Germany infection peaks on Mar 31, 2020 and a total of 187.73 [187.69 - 187.78] thousand people will be infected.
As of Jun 16, 2020, the model predicts that Spain infection peaks on Mar 30, 2020 and a total of 244.15 [244.11 - 244.21] thousand people will be infected.
As of Jun 16, 2020, the model predicts that US infection rate will peak on Apr 13, 2020 and a total of 2114.08 [2114.03 - 2114.15] thousand people will have been infected.
As of Jun 16, 2020, the model predicts that Japan infection rate will peak on Apr 20, 2020 and a total of 18.43 [17.90 - 18.98] thousand people will have been infected.
As of Jun 16, 2020, the model predicts that UK infection rate will reach its peak on Apr 16, 2020 and a total of 298.36 [298.32 - 298.41] thousand people will be infected.
As of Jun 16, 2020, the model predicts that Netherlands infection rate will reach its peak on Apr 07, 2020 and a total of 49.20 [49.16 - 49.26] thousand people will be infected.
I got the idea for this site from a reddit post and the Github repo of Jeroen Kools.
The data all comes from the JHU CSSE Github repo and is updated every day.
I also stole a bunch of the code from Ethan Kruse’s Github repo, which is a fork of one from Elisa Quintana, where he worked out how to make Bokeh and Github Pages play nicely together.
The site is built with Github Pages, Bokeh, PyMC3, and other code.