AttributeError: 'GCNConv' object has no attribute 'lin' when using GCN_2l

[matekenya]

I am attempting to use the predefined models in the framework to perform a classification task and explanation on graphs. Upon initializing the models, using;

model = GCN_2l(model_level='graph', dim_node=num_node_features, dim_hidden=64, num_classes=num_classes)

The following error comes up. AttributeError: 'GCNConv' object has no attribute 'lin'

Previously I was met with the error AttributeError: 'GCNConv' object has no attribute 'weight' when I ran the following; explainer = SubgraphX(model=model_gcn_2l, num_classes=num_classes, device=device, explain_graph=True) explainer(x, edge_index)

It was recommended that I change my torch_geometric version to 1.7.2, which I did now, I am meeting this error AttributeError: 'GCNConv' object has no attribute 'lin' when initializing the model.

[CM-BF]

Hi, we have updated our version to support PyG 2.0.x. If you are using the dig-stable branch, you can use PyG 2.0.x. Sorry for the inconvenience, the incompatibility of PyG is a troublesome problem. You may refer to https://github.com/divelab/DIG/blob/dig-stable/examples/xgraph/gnnexplainer.ipynb.

[matekenya]

This is helpful, thanks.