Source code and datasets for training several Graph Neural Network (GNN) architectures for the task of graph classification in the 3 settings described in the main paper: original, 2stg, and 2stg+

In this work, we compare the performance of each GNN architecture on the graph classification task when the architecture is trained in the following 3 settings

• Original
• 2stg
• 2stg+ Details of these 3 settings are described in the main paper.

The packages and libraries that were used to run the code are:

torch 1.7.0
networkx 2.5
sklearn 0.23.2
numpy 1.16.4
torch-geometric 1.16.3

Code
  |__eigengcn
  |__sag
  |__sage+gat+diffpool

The commands to run to train each GNN architecture on the 3 settings (original, 2stg, 2stg+) are described in the 'run_examples.txt' file in the respective folder.

The source code in this directory is licensed under the MIT license, which can be found in the LICENSE file.