I find it a little unintuitive and inconvenient that BatchedDGLGraph.local_var()
does not return a BatchedDGLGraph
object. Both local_var
and local_scope
can be used to avoid accidentally overriding existing feature data. But I prefer using local_var
because using local_scope
needs one extra indentation. But since local_var
does not return a BatchedDGLGraph
object, I cannot perform operations like broadcast_nodes
. Could BatchedDGLGraph.local_var()
be changed to return a BatchedDGLGraph
object? What’s the reason for the current implementation?
Hi, thanks for your report and I agree with you. This should be fixed in PR 1146. If urgent, you may install from source and modify your local code directly.