[GraphBolt][PyG] Support for Heterogeneous Graphs in to_pyg_data

[Tellili]

Description:

The to_pyg_data method currently only supports homogeneous graphs and a single feature. It would be great if this method could also handle heterogeneous graphs and support multiple features.

Request:

Add support for heterogeneous graphs.
Allow multiple features in the method.

[mfbalin]

@Tellili You might want to look at our advanced example and see how you can use sampled_subgraphs directly for increased performance for the homogenous case. It is also possible to use the same approach for the heterogenous case.

minibatch.sampled_subgraphs is the data source. I am sure it is possible to use PyG in the hetero case with it as well. Our to_pyg_data is currently not recommended for best performance but for convenience with existing PyG models.

[mfbalin]

I might also modify the convert_to_pyg function in our advanced PyG example, make it compatible with hetero PyG use cases and add it to be part of GraphBolt after adding a hetero example.

[Tellili]

Thank you, @mfbalin! I will look into this and look forward to seeing your example.

[mfbalin]

@Tellili Feel free to look at #7722 to see how PyG can be used with GraphBolt for heterogenous graphs. I am working on polishing and verifying if things work properly.

[mfbalin]

#7744 is an alternative, more performative way to convert to a format consumable by the PyG GNN layers.