Option to return only specific properties from VectorRetriever.search results

[oskarhane]

To not return the full nodes on plain vector search for a faster and leaner retrieval and minimize the needed post processing.

Usage:

index_name = "movie_plots"
pluck = ["title", "plot"]

retriever = VectorRetriever(driver, index_name, custom_embeddings, pluck=pluck)

query_text = "What's a movie about a wedding?"
results = retriever.search(query_text=query_text, top_k=2)
# [
#   Neo4jRecord(node={'title': 'The Wedding Ringer', 'plot': 'Two weeks shy of his wedding, a socially awkward guy enters into a charade by hiring the owner of a company that provides best men for grooms in need.'}, score=0.9319988489151001),
#   Neo4jRecord(node={'title': "My Best Friend's Wedding", 'plot': "When a woman's long-time friend says he's engaged, she realizes she loves him herself... and sets out to get him, with only days before the wedding."}, score=0.9314284324645996)
# ]

[willtai]

I'm curious why do we call it pluck? I'm wondering if something like return_node_properties or return_properties is more descriptive

[oskarhane]

I'm curious why do we call it pluck? I'm wondering if something like return_node_properties or return_properties is more descriptive

It comes from https://lodash.com/docs/3.10.1#pluck which is a very popular utility library. Maybe there 's a better word for it, but pluck is pretty descriptive. Maybe pluck_properties is better.

return_* don't sit well with me because there's no exclusivity in the name. return_only_* would be better but that's too long and basically what pluck means.

pick_properties might be an option.

[oskarhane]

After some consideration, I agree with you. Let's go with return_properties @willtai @stellasia

[willtai]

LGTM 👍