graph representation: what should we pass around

choderalab / pinot

Probabilistic Inference for NOvel Therapeutics

MIT License

15 stars 2 forks source link

graph representation: what should we pass around #40

Open yuanqing-wang opened 4 years ago

yuanqing-wang commented 4 years ago

we decided that for now we are going to pass only features in our message passing framework. this is safer but might add complexity in the future when we also wanted to have heterograph

yuanqing-wang commented 4 years ago

looks like it's recommended by dgl to pass only tensors around and we haven't yet find a way to efficiently copy the graph.

note that calling g.local_var() doesn't copy the graph