Added new layer that can handle temporal heterogeneous graphs: HeteroGCLSTM that works similar as GCLSTM
Test for new layer
Documentation for new layer
I couldn't find any literature on layers for temporal graph machine learning for heterogeneous graphs. The intention is to model dictionaries for all node types (keys) with hidden/cell state matrices (values) instead of a single matrix as it is the case for homogeneous graphs. As PyG's ChebConv is not compatible with HeteroData objects I changed it with SAGEConv (had a talk with mathematicians working on Graph ML at our university and they think that shouldn't make too much of a difference). This layer works for the new heterogeneous data structure for now and could be improved in the future.
Are you okay with that @benedekrozemberczki ?
benedekrozemberczki
commented
2 years ago
HeteroGCLSTMthat works similar asGCLSTMI couldn't find any literature on layers for temporal graph machine learning for heterogeneous graphs. The intention is to model dictionaries for all node types (keys) with hidden/cell state matrices (values) instead of a single matrix as it is the case for homogeneous graphs. As PyG's
ChebConvis not compatible withHeteroDataobjects I changed it withSAGEConv(had a talk with mathematicians working on Graph ML at our university and they think that shouldn't make too much of a difference). This layer works for the new heterogeneous data structure for now and could be improved in the future. Are you okay with that @benedekrozemberczki ?