This repository is a COVID-19 tracker that performs some extra analysis on top of the existing, published case data. The tracker scrapes data from the following sources:
There are a few different trackers out there, such as, in no particular order:
However, these are, for the most part, reporting the raw case counts. This is useful information but can miss some important context. This includes things like the lag-time inherent to the testing.
For example, assume that a COVID-19 test takes between one to three days to get a result (postive or negative). Let's also assume that the true number of confirmed cases looks like this, where doubling takes two days:
| Day | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
|---|---|---|---|---|---|---|---|---|
| Count | 1 | 1 | 2 | 2 | 4 | 4 | 8 | 8 |
| New | 1 | 0 | 1 | 0 | 2 | 0 | 4 | 0 |
However, if it takes a test between one to three days to get a result, then what you observe will be:
| Day | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 |
|---|---|---|---|---|---|---|---|---|---|
| Count | 0 | 1 | 1 | 1 | 1 | 4 | 4 | 6 | 8 |
| New | 0 | 1 | 0 | 0 | 0 | 4 | 0 | 2 | 2 |
There appears to be a spike of four cases on day '6', even though the growth rate hasn't changed. Basically, you need to look at the number of cases over the expected lag in order to get a true sense of where things are. Otherwise, you have an extended period where things look stable but actually aren't.
A lot of this was inspired by these two YouTube videos:
They provide a good overview of the underlying mathematics behind an outbreak and exponential growth in general.
The best way to set up the development environment is with Conda. case-rate is meant to run using a GitHub Actions workflow and the commands below have not been tested on Windows.
Create and activate the environment with
$ conda env create
$ conda activate case-rateBuilding the web interface requires NodeJS 16.x. Install the node dependencies with
$ npm ciBuild the reports with
$ make reportThe HTML report will be in the dist folder.
There is also an example Jypter notebook ([algorithm.ipynb]) that shows how the time series prediction works. Running it locally will require installing some extra packages into your environment, i.e.,
$ pip install jupyter matplotlib ipympl