Length of the resulting array for each node is 451 according to:
MATCH (n:Host)
WITH n limit 1
RETURN length(n.embedding)
That seems quite a lot with only 2 iterations and thus calculating cosine similarity for 8k nodes can take 5+ minutes on a 16gb and the whole process on a big graph would be quite slow for now.
Using same data as in https://github.com/neo4j-graph-analytics/ml-models/issues/6 data: https://snap.stanford.edu/data/p2p-Gnutella09.html
import:
LOAD CSV FROM "file:///p2p-Gnutella09.txt" as row fieldterminator ' ' WITH row SKIP 4 MERGE (h1:Host{id:row[0]}) MERGE (h2:Host{id:row[1]}) MERGE (h1)-[:CONNECTION]->(h2)
Run algo with 2 iterations and pagerank node property:
Length of the resulting array for each node is 451 according to:
That seems quite a lot with only 2 iterations and thus calculating cosine similarity for 8k nodes can take 5+ minutes on a 16gb and the whole process on a big graph would be quite slow for now.