Spectral Graph Convolution Experiments

For implementations of SignNet and BasisNet, see signbasisnet.py and training.py, specifically the functions gen_sign_inv and gen_basis_inv.

To run the experiments, use the scripts in scripts/.

To run our SignNet and BasisNet, use

bash scripts/sign_basis_inv.sh

You can also pass in a filter type (one of low, high, band, rejection, comb) e.g.

bash scripts/sign_basis_inv.sh rejection

Attribution

These codes are built off of the [BernNet repo] by He et al. in 2021, which in turn builds off of experimental setups in [this repo] from "Analyzing the Expressive Power of Graph Neural Networks in a Spectral Perspective" by Baliclar et al. 2021.

The Invariant Graph Networks (IGN) implementation in Pytorch is taken from [this repo] by Hy Truong Son.