kedar2 / Oversquashing

Oversquashing in GNNs through the lens of information contraction and graph expansion

Oversquashing in GNNs through the lens of information contraction and graph expansion

Code for the synthetic experiments in our paper on oversquashing in GNNs to appear in the 58th Annual Allerton Conference on Communication, Control, and Computing (preprint: arXiv:2208.03471).

Requirements

To configure and activate the conda environment for this repository, run

conda env create -f environment.yml
conda activate oversquashing

Spectral expansion plots

To test the spectral expansion of a graph under RLEF, G-RLEF, or SDRF, run the file plots.py. By default, it will produce a figure for a path of 3 cliques, each with 10 nodes.

NeighborsMatch experiment

To run the NeighborsMatch experiment, run the file run_neighborsmatch.py. The following commands will run the experiment for the G-RLEF and SDRF rewirings.

python run_neighborsmatch.py --rewiring GRLEF
python run_neighborsmatch.py --rewiring SDRF

The full list of settings can be found in the file hyperparams.py.