Deep Learning-Derived Optimal Aviation Strategy to Quench Pandemics

Official PyTorch implementation of DCSAGE and DCGAT from Deep Learning-Derived Optimal Aviation Strategy to Quench Pandemics.

Training

Getting Started

This codebase was developed using Python 3.8.12 and PyTorch 1.8.0. To reproduce the environment, install dependencies through either anaconda (preferred) or pip (just use pip install command):

conda env create -n DCSAGE python=3.8
conda activate DCSAGE
conda install pytorch==1.8.0 -c pytorch
pip install -r requirements.txt

and then set the PYTHONPATH to the base directory of the repository (important for imports to work correctly):

export PYTHONPATH="/path/to/base/DCSAGE/directory" 

DCSAGE Training

To train a single DCSAGE model, run:

cd training
python train.py

For training multiple DCSAGE models, define the number of models (e.g. 100) in training/train_multiple_models.py and run:

cd training
python train_multiple_models.py

DCSAGE Node Perturbation

To run the node perturbation experiment on trained DCSAGE models, first update the directory of the trained models in ./explainability/node_perturbation/node_perturb_analysis_config.json. Then run:

cd explainability/node_perturbation
python node_perturbation_analysis.py