README.md

Improving Expressive Power of Spectral Graph Neural Networks with Eigenvalue Correction

This code contains a PyTorch implementation of "Improving Expressive Power of Spectral Graph Neural Networks with Eigenvalue Correction" (AAAI 2024)

Environment Settings

• pytorch 1.8.1
• numpy 1.18.1
• torch-geometric 1.6.3
• tqdm 4.59.0
• scipy 1.6.2
• seaborn 0.11.1
• scikit-learn 0.24.1

Histogram of eigenvalue distribution in Figure 1 in the paper

You can run the following Command:

python histogram.py

The number of distinct eigenvalues in Table 1 in the paper

You can run the following Command:

python distinct_eigvalues.py

Eigendecomposition

Run the following command to perform eigendecomposition on all ten datasets, and will print the time required for each eigendecomposition.

python eigendecomposition.py

The eigenvalues and eigenvectors are stored in the data directory.

Running the code

You can run the following script directly:

sh EC-Bern.sh

or run the following Command

sh EC-GPR.sh