How does graph partition function with random partition method work?

Questions
#1

Great library! I am currently learning distributed training of GNN with DGL. I have tested both metis and random partition methods and surprisingly found out that the random partition method always returns a graph partition with 0 edge cut (ogbn-proteins/ogbn-products dataset with 10000 partitions). It seems the random partition only assigns each node to different partitions with uniform probability, so I am confused as to why the edge cut is still 0.

Any help would be appreciated!

#2

How did you got the edge cut between different partition?

#3

The edge cut value is returned by dgl.distributed.partition_graph function. Something like this:

image
image1832×954 203 KB

#4

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.