Project developed for the Geometrical Data Analysis course of the MVA Master (ENS Paris-Saclay) by:
- Alice Valença De Lorenci
- Julián Avalarez de Giorgi
The goal of the project was to study the following paper: RonLevie,FedericoMonti,XavierBresson,andMichaelM.Bronstein. 2017. CayleyNets: Graph Convolutional Neural Networks with Complex Rational Spectral Filters. CoRR abs/1705.07664 (2017). arXiv:1705.07664 http://arxiv.org/abs/1705.07664
The repository contains the source code associated with the project, notably, we propose a Pytorch Geometric implementation of Cayley convolutional layers, the building block of CayleyNets.
The following package versions were used:
numpy 1.26.2
torch 2.0.1+cu117
torch_geometric 2.4.0
matplotlib 3.5.1
scipy 1.8.0
sknetwork 0.31.0
pickle 4.0
This repository is organized as follows:
- src/CayleyNet.py: implementation of Cayley convolutional layers and CayleyNets
- src/CayleyTransform.py: implementation of the Cayley transform
- src/ChebNet.py: implementation of ChebNets
- src/CommunitiesGraph.py: communities graph handler
- src/CORA.py: CORA dataset handler
- src/Dataset.py: generic dataset handler
- src/utils.py: miscellaneous helper methods used to train and evaluate the models
- CayleyTransform.ipynb: experiments around the Cayley transform
- Experiments.ipynb: experiments with Spectral GNNs