This repository was numerical basis of the masters thesis "On the Transferability of Spectral Graph Neural Networks". To find the final Master's thesis: https://infoscience.epfl.ch/record/278148?&ln=fr
Regroupment of functions used in all of the project
lib/
├── dataset.py
├── graphs.py
├── laplacian.py
├── model.py
├── ninja.py
└── utils.py
Shared resources between different part of the project
- models regroup several trained ChebNets that are all supposedly achieving 98% accuracy over their testing set
├── imgs
├── models
Notebook implementation of the Chebyconv using the DGL framework. The net is a simple LeNet 5 with three convolution layer and two linear layers. The dataset is the good-old MNIST
├──ChebGCNs.ipynb
├── README.md
├── Eigonvalues_Viz.ipynb
├── Gui.py
├── Training_Chebnet_on_MNIST.ipynb
├── filtervisualisation.ipynb
├── laplacianprediction.ipynb
├── transform_training.ipynb
└── visualisations.ipynb
Notebook for stochastic learning. For now only two notebooks to train and test the model are implemented.
├── Training
├── Testing