SMCDC2021

Repository for our submission to the Smoky Mountains Computational Sciences and Engineering Conference Data Challenge 2021 - Finding Novel Links in COVID-19 Knowledge Graph.

Setup

Note that this repository was created in Python 3.7.10 and all GCN models were trained on a NVIDIA TITAN V GPU. To set up, perform the following:

1. git clone https://github.com/RemingtonKim/SMCDC2021.git
2. cd SMCDC2021
3. pip install -r requirements.txt
4. Download necessary data from here and store in ./data.

Code Description

The following describes the purpose of the files in ./src.

File	Description
preprocessing.ipynb	Cleans raw data and preprocesses it into networkX and PyTorch compatible formats.
analysis.ipynb	Performs various network analyses.
baselines.ipynb	Runs all link and leadtime prediction heuristic and node2vec baselines.
models.py	Contains all the GCN models built in PyTorch.
utils.py	Contains helper functions.
train_*_*.py	Performs training, validation, and testing for a GCN model.

Note that preprocessing.ipynb should be run before anything else as its outputs are used in the other notebooks and files. To run a GCN model, run python3 train_*_*.py. Set MODE = 'test' if the model has been trained. .pth and log files for trained models will show up in ./models.

Contact

• Remington Kim (remingtonskim AT gmail.com)
• Yue Ning (yue.ning AT stevens.edu)