edge attributes in heterogeneous graphs?

[paapu88]

Dear All,

I have been reading this nice doc page, and was wondering:

Are there any examples around where edge attributes would have been used when predicting/classifying with heterogeneous graphs?

Terveisin, Markus

[rusty1s]

We currently do not have such an example as best as I know, but edge features are well supported in heterogeneous graphs/heterogeneous GNNs in PyG. If you run into any trouble to integrate, I am happy to help.

[paapu88]

Hmm, for instance here
It seems to me that in forward there are no edge attributes.
Or have I misunderstood this?

[rusty1s]

Yes, HGTConv does not support edge features, but, e.g., HeteroConv or to_hetero does.

[paapu88]

Thanks!
I got confused by this page at block:

model = HeteroGNN(...)
output = model(data.x_dict, data.edge_index_dict, data.edge_attr_dict)

Because one gets the idea that there is some straightforward class 'HeteroGNN' which can digest 'data.edge_attr_dict'

Hope to have got it now. One might consider re-writing the documentation to make it more clear for beginners...