The difference from edge-level tasks, is that it not only needs to make prediction of the edge label, but also have to decide if the edge exists at all. Link_pred: Link prediction tasks makes prediction on existance of links (edges). The prediction of each edge is made based on a pair of node embeddings corresponding to the endpoints of the edge. The prediction of each node is made based on node embeddings output by a GNN.Įdge: Edge-level tasks makes prediction on labels for edges. Node: Node-level tasks makes prediction on labels for nodes.
We classify the learning tasks into the following categories, all of which are fully supported by DeepSNAP.īoth classification and regression objectives can be applied to all of the tasks. We first explain some preliminaries for learning on graphs.
0 kommentar(er)
