This tool helps visualize the sum of node features in each layer of a GNN-based model. It can work with random or molecular graphs based on SMILES strings.
To test with random graphs, use the 'GNN'
model_mode.
If you have your own model, import your model class and load your trained model, ensuring that all GNN layers are grouped in self.convs = torch.nn.ModuleList()
.
After the first layer, nearly all nodes have a large node feature sum | while only a few nodes have a large feature sum at the last layer. |
---|---|
The trained CBS model is intentionally not included in the repository and may be added later for academic purposes.