To replicate the experiments in the paper:

Step 0: Environment and Prerequisites

Run the following commands to clone this repo and create the Conda environment:

git clone https://github.com/IanShi1996/PAN-cODE.git
cd PAN-cODE
conda env create -f environment.yml
conda activate covidforecast

Step 1: Obtaining the Data

Update timestamps as appropriate in lib/Constants.py, then run:

python get_country_data.py

Step 2: Model Training

Train the PAN-cODE model using train.py. Hyperparameters can be specified using command line arguments, as documented in train.py.

Step 3: Model Evaluation

The summarize.py script can be used to generate forecasts for integration with the COVID-19 forecast hub.