I am trying to run the node2vec example on two bipartite graphs (with same node2vec configurations):
Graph A:
18M nodes
30M edges
Graph B:
7M nodes
13M edges
However, loading graph A is a lot faster than graph B (~8x), and a lot less memory intensive ( 17 TB vs 124 TB ) [memory data for ProbTransitionMatrix from PredictMemoryRequirements in biasedrandomwalk.cpp]
What is happening under the hood and how can I improve the performance?
I am trying to run the node2vec example on two bipartite graphs (with same node2vec configurations):
Graph A: 18M nodes 30M edges
Graph B: 7M nodes 13M edges
However, loading graph A is a lot faster than graph B (~8x), and a lot less memory intensive ( 17 TB vs 124 TB ) [memory data for ProbTransitionMatrix from PredictMemoryRequirements in biasedrandomwalk.cpp]
What is happening under the hood and how can I improve the performance?