DenseGCNConv with to_hetero

[wgeul]

Hi,

Is it possible to convert DenseGCNConv or other layers that have a forward function that only accepts x: Tensor by default to a heterogeneous implementation? Is there by any chance a quick solution to support x_src and x_dest for such pre-defined layers?

I'm trying to make a model that performs a Graph-level regression task on a heterogeneous graph network and think the Dense layers are most appropriate for the task.

[Padarn]

Does https://github.com/pyg-team/pytorch_geometric/blob/master/torch_geometric/nn/to_hetero_transformer.py help solve your problem?

It allows you to build a heterogenous model out of homogenous modules. There is an example of it's use here

pytorch_geometric/examples/hetero/to_hetero_mag.py

Line 49 in 8bd9ae4

model = to_hetero(model, data.metadata(), aggr='sum').to(device)

[rusty1s]

It is currently not possible, I am sorry. Even for graph-level tasks, we recommend to use sparse representations.

[Padarn]

+1, sorry I originally misunderstood the question

[wgeul]

Thanks for the answers.