Expressiveness of Line Graph Neural Networks

Introduction

Welcome to this project, in which I researched the expressiveness of Line Graph Neural Networks! Please find the project report here. With the code in this repository, I obtained the results of the experiment at the end of the report.

Structure

Files

File	Description
code/main.ipynb	The main results displayed in the report, as well as the used graph figures, can be found in code/main.ipynb. To run this notebook yourself, follow the procedure in the section 'Running code/main.ipynb' of this file.
code/dataset.py	The code in this file was used to preprocess the datasets used in these experiments (i.e. sample subgraphs for link prediction).
code/experiment.py	The code in this file was used to train the models.
code/models.py	This file contains the model implementations.
code/utils.py	This file contains utility functions for the rest of the project.

Folders

Folder	Description
code/study	Contains the TensorBoard logs of the three experiments (subfolder names: PPI-hidden-dim-20, PPI-hidden-dim-52, TwitchEN-hidden-dim-20).
code/data	Is supposed to contain the dataset. If you want to run code/main.ipynb yourself, you will need to fill this folder with the right data, as outlined in the next section.
report	Contains the TeX code used to generate the report.

Running code/main.ipynb

To run code/main.ipynb, you will need to download the datasets from the following link: https://drive.google.com/drive/folders/1KiYGXAuR-3VBO31yu82S8QLrxGLgwc_2?usp=sharing and create the following folder structure in the code/data folder. Then, everything should run!

| data
|---- TwitchENDataset
|---- PPIDataset