Over-Squashing

Figure from "On Over-Squashing in Message Passing Neural Networks" by Francesco Di Giovanni et al

Overview

This repo contains code for the experimental part of the paper "FSW-GNN: A Bi-Lipschitz WL-Equivalent Graph Neural Network".

Over-squashing is a phenomenon in Graph Neural Networks (GNNs) where information bottlenecks hinder effective message passing over long-range dependencies. This project implements mechanisms to mitigate over-squashing, improving GNN performance on tasks requiring deep Networks. It is built using Python and deep learning frameworks, including PyTorch and PyTorch Geometric.

Installation

To set up the project environment and install all necessary dependencies, follow these steps:

git clone https://github.com/yonatansverdlov/Over-squashing.git
cd Over-squashing
conda env create -f dependencies.yaml
conda activate oversquash

Usage

This project includes three types of experiments: Over-Squashing Experiments, Transductive Learning, Graphh Classification, and MolHIV & LRGB.

Over-Squashing Experiments

Navigate to the bottleneck directory:

cd bottleneck

Choose a data_type from the following options: Ring, Tree, CrossRing, CliquePath.

For Tree experiments:

python train.py --task_type Tree --min_radius 2 --max_radius 9

For Ring, CrossRing, or CliquePath:

python train.py --task_type data_type --min_radius 2 --max_radius 16

Transductive Learning

Navigate to the bottleneck directory:

cd bottleneck

Choose a data_type from the following options: Cora, Cite, Pubm, Cham, Squi, Actor, Corn, Texas, Wisc.
Set repeat to a value between 1 and 10 to determine the number of different seeds:

python train.py --task_type data_type --repeat repeat

Graph Classification (Mutag & Protein)

For MUTAG and Protein datasets, specify the number of different seeds (repeat) and run:

python train.py --task_type data_type --repeat repeat --min_radius 3 --max_radius 4

For LRGB and MolHIV:

cd Over-squashing
conda create --name lrgb -c conda-forge python=3.10
conda activate lrgb
pip install -r lrgb_requirements.txt

License

This project is licensed under the MIT License.

Contact

For any questions or feedback, reach out to me at yonatans@campus.technion.ac.il.