I created a big graph database from neo4j . There are 37 million nodes, 161 million edges. I am wondering if dgl library can handle this giant graph. Or do I have to sample the giant graph to create a small sub-graph before importing it into dgl for deep learning ?
Have you tried creating a
DGLGraph from the list of edges? You can sample subgraphs during training.
Hi, can you share your hardware information? We have already trained 500M nodes graph in AWS EC2 x1.x32large instance. You can see our example code in