GraphSAGE
GraphSAGE copied to clipboard
Some problem of evaluating on new test data
Hi, does this model can be used to unseen nodes if we don't know adj_info of these nodes? That's mean, we only know the features of these nodes.
It won't work out-of-the-box. If we do not know the connectivity of new nodes, the network won't know how to propagate information from it's neighborhood. But I think if in your dataset, features of nodes are related to its connectivity, maybe you can predict potential neighborhood and then use GraphSAGE on the predicted neighborhood.
Can GraphSAGE represent the new graph that is not connected with the trained graph? @RexYing
What do you mean about the adj_info. Does it include the neighborhoods and features of neighborhoods?