Batch same structured graph

Questions
#1

Q1. If my graph structure is steady, it’s only the node features different in each graph. Do I need to batch the graphs using BatchedDGLGraph? Because in the tutorial, they are all graphs with different structures. Just wondering if there is a more efficient way.

Q2. If there is no edge information and I want to use the attention weights to represent the edge connections. Do i just first assume every nodes are connected to each other, and create edge feature ‘w’ as the weights and use attention to update the weights?

Efficient way for custom non-graph dataset
#2

Q1. If you deal with one graph at a time then you can use a single DGLGraph object and only replace features when necessary. If you want to deal with multiple graphs at a time, then yes.

Q2. Yes. When you do not have information of the connectivity, then you can first treat it as a completely connected graph and then adaptively learn attention weights.

#3

Thanks for your reply, that clarifies my doubts!