I am trying to use metis on reddit datatset, and I get reddit.json.
Question 1: whether node_map means the node allocation to different parts? As is shown below, for example, I think node id 0-45315 have assigned to part0?
Question 2: But when I try to load the graph.dgl on part0, the node num is up to 100000, why?
“node_map”: {
“_N”: [
[
0,
45315
],
[
45315,
91691
],
[
91691,
139560
],
[
139560,
186931
],
[
186931,
232965
]
]
Q1
Yes.
Q2
It’s due to HALO nodes. If you have loaded graph.dgl, you can check if nodes belong to current partition or not via checking g.ndata['inner_node'].
So for Q1，node_map means that that node features which have allocated to different parts? And Q2 means it stores graph structure but not include node features?
yes. feat data are stored in node_feat.dgl while graph structure is in graph.dgl.
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.