Scalability and optimization

Handling large-scale graphs requires advanced techniques for efficient training and inference. We covered the challenge of scalability in Chapter 5; here, we’ll look at practical examples of techniques that can help us address this issue.

Mini-batch training with neighborhood sampling

Instead of processing the entire graph, we can use mini-batch training with neighborhood sampling:

1. Sample a subset of user nodes:

   def create_mini_batch(G, batch_size, n_pos=5, n_neg=5, 
                         n_neighbors=10, n_hops=2):
       # Get all user nodes
       all_user_nodes = [n for n in G.nodes() if 
                         n.startswith('user_')]
       if not all_user_nodes...