[GNN-Explainer] Model can't be interpreted

[zju-wyh]

I create a model including CNN and GAT,but it can't be interpreted using GNN-Explainer.Can it only explain the GNN model not including CNN and Linear？Looking forward to your answer.

[rusty1s]

To the best of my knowledge that should work as long as you have a single message passing layer in your network. Do you have a minimal example to reproduce your issue?

[zju-wyh]

Thanks for your answer. Model only with GNN is ok, but adding convolutional layer and Linear layer will report an error:TypeError: forward() got an unexpected keyword argument 'x'.

This is ok

This is my model

[rusty1s]

This is mainly because your forward function should not expect data as input but rather x and edge_index (and any additional arguments).

[HeVLF]

Hey @rusty1s
Can my forward function receive x, edge_index, and edge_attr or, for the GNNExplainer to work, the forward function should only receive x and edge_index?

[rusty1s]

The GNNExplainer can take in any arguments that the forward pass requests. It is just that x and edge_index needs to be part of it in order to correctly set up the gradient computation.