lkmuk/COVID19-Dashboard

An interactive dashboard application based on plotly/Dash framework. Featuring a primitive SIR predictive model. Additionally, the app allows us to analysis how well the countries of interest are coping with the pandemic at various points of time, thereby enabling us to study how certain public health measures might affect the pandemic spread.

An Interactive COVID-19 Dashboard

Functionalities

An interactive GUI based on Dash/ Plotly framework that:

• before the dashboard application initializes: automatically fetches (updates) & process¹ the raw data from the COVID-19 Repo published by CSSE, John Hopkins University;

• allows user to select those countries present in the dataset;

• for each selected countries:

  • in part 1 of the dashboard: plots the respective time series of the cumulative number of confirmed cases (in the first interactive figure);

  • in part 2 of the dashboard:

    • plots the empirical time series of the number of infected people (in the second interactive figure);

    • allows user to specify the observation window. The SIR parameters during that period is identified using Least-Square criterion (in other words, a SIR model is "fitted" to the data recorded in this period);

    • simulates the time series & plots it on the second interactive figure. The predictive segment is a 30-day period following the observation period;

    • presents those SIR parameters in an interactive table;

    • shows some diagnostic messages to the user.

A quick demo can be found in the first 3 minutes of this video:

app_basics and app_SIR_country correspond to parts 1 and 2 of app_full respectively.

Setting up the project

Below are the steps required for running the application (any one of src/app_full.py, src/app_basics.py or src/app_SIR_country.py)

0. Before beginning the setup, ensure your computer has git (for proper cloning of the repository & updating the raw csv data in form of git submodule dependency), python 3 and pip3.

1. Set up your local repository in your desired directory.

   $ git clone --recurse-submodules https://github.com/lkmuk/COVID19-Dashboard.git
   $ cd COVID19-Dashboard/data_raw/COVID-19
   $ git checkout master

2. Set up the virtual environment. (continued from above)

   $ cd ../..
   $ python3 -m venv venv
   $ source venv/bin/activate
   $ pip3 install -r dependencies.txt

3. Run any one of the three applications. This can be, for example: (continued from above)

   $ cd src
   $ python3 app_full.py

   note: because of the uses of relative paths inside the applications app_<full/...>.py for accessing both the raw and processed data, please call the application only when you are currently in the directory of src.

If you also want to execute the test scripts in the source folder and the notebooks, please also install also matplotlib and jupyter notebook respectively in the virtual environment.

To-do/ Potential enhancements

This apparently, concerns only part 2.

• improve the numerical accuracy of the identification parameter/ more rigorous analysis into this issue
  • smooth the empirical SIR traces before any parameter identification?
  • more sophisticated version to est_varying_BetaGamma?
  • ways to compensate for measurement delay?
  • more advanced epidemic models than SIR?
• create more plausibility check rule, to guard against erroneous identified parameter.
• more memorization technique to reduce callback latency. So far, I've only eliminated the computationally expensive "refitting" when only some minor display options are changed, e.g. normalizing X-axis. (see the console message of app_full.py) In the future, it is advantageous to refit only if
  • the observation window is changed, OR
  • a new country is added to the selection. (now we "refit" the parameters whenever the selection list changes OR the obs. window spec. changes,)
• ...

Footnotes

1. specifically, aggregates the time series data by country, and perform some arithmetics to yield the empirical time series of the number of infected and "removed" people for each country) ↩