Hide nodes in a graph

[riskiem]

I have about 20 graphs, with 16-4 test-train split. What I want to do is within each graph hide a few nodes and then perform classification. Is there a way to do this?

[rusty1s]

You can use a transformation which hides nodes via utils.subgraph:

def hide_random_nodes(data):
    idx = torch.randperm(data.num_nodes)[:10]  # Use 10 random nodes
    data.x = data.x[idx]
    data.edge_index, data.edge_attr = subgraph(data.edge_index, data.edge_attr, relabel_nodes=True)
    return data

TUDataset(root, name='MUTAG', transform=hide_random_nodes)