This project contains a reimplementation and extension of [1] using PyTorch and FrEIA. Concretely, we apply the BayesFlow model architecture to estimate parameters of an epidemiological model for simulating COVID-19.
- Setup a new virtual environment (we recommend using Anaconda
- Install FrEIA using
pip:pip install git+https://github.com/VLL-HD/FrEIA.gitor check out the documentation. If you want to use GPU, make sure your PyTorch installation is properly set up for CUDA. - Install the other requirements in
requirements.txt
For training a new model, simply run python main.py. For an overview over parameters, run python main.py --help.
Our results are in the provided jupyter notebooks. You can have a look at them. The code for evaluation is taken from https://github.com/stefanradev93/AIAgainstCorona and modified for our purposes.
[1] Radev, S.T., Graw, F., Chen, S., Mutters, N.T., Eichel, V.M., Bärnighausen, T., Köthe, U.: Model-based bayesian inference of disease outbreak dynamics with invertible neural networks. arXiv preprint arXiv:2010.00300 (2020)