Error with starting trainer for large graphs

Questions
#1

Hi,

I’m preparing DGL to train some large graphs (in excess of 10,000 nodes). As a pilot, I wanted to run the example script mentioned on giant_graphs by using (python run_store_server.py --dataset reddit --num-workers 4). I’m stuck with an error : TypeError: create_graph_store_server() got an unexpected keyword argument ‘edge_dir’.

Could you please let me know if I’m using outdated script?

Thanks so much.

#2

I think for a graph with just about 10, 000 nodes, you don’t really need graph store. Try following the sampling-based training here.

#3

Thanks Mufei … appreciate it.

#4

You may not even need sampling-based training. If you have multiple graphs of about 10,000 nodes, you can train your model by constructing mini-batches of graphs. You may also be interested in our tutorial on graph batching.

#5

Looking at it. Thanks so much.