How to perform message passing in HeteroGraph with empty node,edge types

[DogitoErgoSum]

Hi,
My main goal is to perform HeteroGraph classification. Not all graphs will have all node types and will have different node type combinations, as shown below:

Here L0_S5_F114 has no nodes present.
I tried to perform a simple message passing on a graph like with the following code but it failed.

conv = HeteroConv({edge_type: GraphConv((-1, -1), 5) for edge_type in data.edge_types}, aggr='sum').to(device)

conv(data.x_dict, data.edge_index_dict, edge_weight_dict=data.edge_attr_dict)

What am I doing wrong or what should I be doing?

[DogitoErgoSum]

The problem here is that there is message passing occurring in empty node types, but I need to keep the empty node types to be able to create batches.
How do I perform message passing only on non-empty nodes?

[rusty1s]

Thanks for reporting. I will try to reproduce this on my end. If you also have a script to reproduce, please let me know. Message passing with empty node types should definitely work in PyG >= 2.0.4

[DogitoErgoSum]

The version seems to be the problem, i will check if it fixed.
TY

[rusty1s]

You can simply try to install via pip install torch-geometric==2.0.4. Let me know.

[DogitoErgoSum]

It works now, ty for your fast and accurate response