[[TOC]]
This project aims to quantify the impact that the Covid-19 pandemic has on the global economy, utilising machine learning. One of the main limitations in machine learning is the modelling of unsampled regions in the feature space. Furthermore, models learned during the training process are only valid in the context of unchanged underlying data generation processes. Since the Covid-19 pandemic fundamentally changed the way our society operates, and we don't have any samples of a global pandemic of smaller and/or greater size, it is impossible to directly model the impact of the virus on different branches of the economy. Instead, we aim to model the economy without the pandemic ensuing from January 2020 on. The impact of the pandemic can then be measured as the difference of the predicted courses without Covid-19, and the actual, observed outcomes. The models and its predictions are made accessible with an interactive web interface.
In order to run the project, you need to have Docker and docker-compose installed on your machine. Running the project with Docker takes care of installing the correct project dependencies and environments.
After cloning the directory, you can start the project by running the docker-compose file. This starts both the front- and the backend of our web server. Both docker containers for the front- and backend will build and start up automatically. You can now access the webpage using any browser of your liking at http://localhost:8080/#/.
NOTE: Start the docker-compose from the root directory of the project using:
<user>:~$ docker-compose up
WARNING: Building the docker image for the first time takes a lot of time. Some installation steps are complex, and the libraries are quite large. Even if the installation seems to stop, be patient.
All data sets can be found in the res folder. Most Subsets have their own README explaining particular details for this individual set such as the location of related code or comments on the quality.
NOTE: Not all data sets were used in the final project.
The project documentation can be found in the documentation folder. The project's code is documented with Doxygen. To access the documentation, simply open the index file with a web browser of your liking. The documentation folder also includes all milestone reports as well as some meeting transcripts. The final report is located here.
Our code base is located in the src folder. Scrapers, API scripts etc. to obtain raw data can be found in the data_collection folder. Basic preprocessing and data extraction scripts are located in data_management. The unified data preprocessing scripts are saved in data_pipeline. The code for training, final model evaluation and live inference for the web interface is located in inference. Finally, webinterface contains the code for both front- and backend. As mentioned before, a complete documentation of the code can be accessed by opening the Doxygen documentation in a web browser.
The model parameters were determined by cross validation. We created a dictionary with the optimal parameters in our model file. Using the parameters, the model is trained on the respective time series.