Node Heterogeneous Hypergraph Implementation

[sheyril]

Hi, I'm trying to carry out node classification using HypergraphConv provided by the library. I'm wondering if the implementation will work with heterogeneous nodes (I don't necessarily need to use attention at this point)?

The implemented paper only mentions heterogeneity in the context of attention. From what I see in the docs, the forward function expects me to input a node feature matrix of dimension: (Number of nodes, Number of Features), suggesting homogeneous nodes.

I'm quite new to both Hypergraphs and Heterogeneous data, I'd appreciate any help. Thank you for your time!

[wsad1]

I haven't tested this.
But maybe you could use HeteroConv. For each edge_type you could pass the HypergraphConv.

[rusty1s]

What you can do is store incidence matrices instead of adjacency matrices within HeteroData, e.g., a node type now corresponds to a set of hyperedges. You should then be able to leverage regular GNN layers to process.