5

Has anyone successfully integrated Apache Mahout with Neo4j as a datastore? If so, how much works was involved, and what was the performance like?

DeejUK
  • 12,891
  • 19
  • 89
  • 169

1 Answers1

3

I haven't heard from anyone that's tried it. There is no neo4j-specific connector code in the project.

To the extent you can use neo4j with Hadoop, you can use it with Mahout; Mahout doesn't have a lot to do with particular data stores.

That's the good news, I suppose; there's not much about Mahout that's particularly tied to a data store, so yes with a little work I'm sure you can get something working.

I don't think a graph database is the most natural choice to store the sorts of stuff Mahout crunches, which are just big vectors.

Sean Owen
  • 66,182
  • 23
  • 141
  • 173
  • If Mahout is all about graphs of vertices, then Neo4j should be a good fit. There's an interesting article http://markorodriguez.com/2011/02/18/mysql-vs-neo4j-on-a-large-scale-graph-traversal/ about its speed advantage over MySQL. – DeejUK Oct 12 '11 at 09:44
  • It's not all about graphs of vertices -- what are you referring to? It's almost all vectors and matrices. – Sean Owen Oct 12 '11 at 12:09
  • Sorry, I'm not working on the project directly, so I'm prolly misunderstanding it a little. I've posted to the Neo4j user group. – DeejUK Oct 14 '11 at 12:06