Interim blog setup, rough notes being dumped here at the moment - I'll tidy once I get to a milestone. And sorry about the layout, that'll probably take even longer to get around to... #TODO

llama_index SPARQL Questions

Published on 2023-09-07 by @danja

Immediate goal is to recreate Wey's Graph RAG demo but using a SPARQL store rather than NebulaGraph.

I'm running it as The in-progress (rels() will be called from get_rel_map()).

I've got data going to the store something like it should using graph_store.upsert(). Data coming back from calls to graph_store.get_rel_map() that appears to be more or less the right shape. But an issue with each I would appreciate advice on.

  1. Batching data

    Right now I've set up INSERT queries to add triplets (as RDF triples) one at a time to the store. It's really inefficient because each has to be wrapped in a query skeleton and POSTed over HTTP. I imagine some kind of batch handling is already implemented somewhere, but I couldn't find anything I understood. It isn't essential right now, but if there is something, pointers?

  2. Loops in graph data

    This is a breaking issue for me right now. I'm sure I can sort it out, but a little more knowledge should make that quicker :) So at some point the data is fed through llama_index/llama_index/response_synthesizers/ I get RecursionError: maximum recursion depth exceeded in comparison, so seems likely there's a loop somewhere. Short term, some kind of check+delete should fix that (algorithm hunt time). But more generally, what is the situation with graphs & loops? Is this an underlying feature of the way Graph RAG operates, or is it just an implementation issue?