How to model a dynamic heterogeneous attributed graph dataset in PyG

[kamizhou]

I‘ve noticed there are HeteroData() and TemporalData() in torch_geometric.data , but how could I combine them to handle or create a dynamic heterogeneous directed attributed graph in PyG ?
For example, TypeA node NA1 with attributes VA1 has a TypeM connection EM1 with attributes VM1 with TypeB node NB1 with attributes VB1 in timestamp T1, while TypeB node NB2 with attributes VB2 has a TypeN connection EN2 with attributes VN2 with TypeC node NC3 with attributes VC3 regardless of timestamp.

[rusty1s]

You can add your timestamp to the node types of a HeteroData object and then use, e.g., NeighborLoader to process that information. NeighborLoader supports temporal sampling via input_time and time_attr such that you only sample nodes that exist in time.