Stochastic training

[rc-94]

Hello,

First, thank you for providing this amazing work ! I may be wrong but have the feeling your HTGNN model is not compatible with the DGL stochastic training (https://docs.dgl.ai/tutorials/large/L4_message_passing.html). I was wondering if you plan to release a HTGNN implementation that would be compatible for both full-graph and stochastic training.

Thanks