Meta-repository for code supporting It's PageRank All The Way Down: Simplifying Deep Graph Networks (SDM23).
@inbook{doi:10.1137/1.9781611977653.ch20,
author = {Dominic Jack and Sarah Erfani and Jeffrey Chan and Sutharshan Rajasegarar and Christopher Leckie},
title = {It's PageRank All The Way Down: Simplifying Deep Graph Networks},
booktitle = {Proceedings of the 2023 SIAM International Conference on Data Mining (SDM)},
chapter = {},
pages = {172-180},
doi = {10.1137/1.9781611977653.ch20},
URL = {https://epubs.siam.org/doi/abs/10.1137/1.9781611977653.ch20},
eprint = {https://epubs.siam.org/doi/pdf/10.1137/1.9781611977653.ch20},
abstract = { Abstract First developed to rank website relevance, PageRank has become ubiquitous in many areas of graph machine learning including deep learning. We demonstrate that a number of recently published deep graph neural networks are qualitatively equivalent to shallow networks utilizing Personalized PageRank (PPR), and that their performance improvements over existing PPR implementations can be fully explained by hyperparameter choices. We also show that PPR with these hyperparameters outperform more recently published sophisticated variations of PPR-based graph neural networks, and present efficient implementations that reduce training times and memory requirements while improving scalability. }
}
- ppr-gnn-tf: tensorflow implementation of PPR-MLP and MLP-PPR models, plus our own tensorflow implementations of DAGNN and GCN2 for fair comparison of training times.
- GCNII: minimal modification of the original GCNII torch implementation with SS-GCN2 variant and a pytorch MLP-PPR implementation (CG-GCN2)
- DeeperGNN: minimal modification of the original DAGNN implementation with SS-DAGNN variant
git clone --recurse-submodules https://github.com/jackd/ppr-gnn-sdm23.git
See README.md
s of constituent repositories for individual setups.