GNNExplainer for edge regression task

[antoni4040]

Hello. Can GNNExplainer be used for edge regression tasks? Or is it only for node prediction and graph prediction?

[rusty1s]

GNNExplainer currently only works for node- and graph-level tasks. Our newly added to_captum functionality works for all tasks though, see this discussion.

[antoni4040]

How do I import to_captum? I can't import it from nn or nn.models.

[rusty1s]

You need to install from master:

pip install git+https://github.com/pyg-team/pytorch_geometric.git

It just got recently added to PyG.

[antoni4040]

I'm trying to do something like this:

edge_index = 1
captum_model = to_captum(model, mask_type='node') 
ig = IntegratedGradients(captum_model)
data1 = geometricData[0]
ig_attr_node = ig.attribute(data1)

And the model's forward function starts like this:

def forward(self, data):
        x, edge_index = data.x, data.edge_index
        x = self.first(x).relu()
        x = self.second(x).relu()
        x = self.third(x)

And I get:

AssertionError: `inputs` must have type torch.Tensor but <class 'torch_geometric.data.data.Data'> found: 

Do I need to change it to accept x, edge_index instead of the bundled Data object?

[rusty1s]

Yes, this is indeed necessary. See here for an example.