Help wrapping my head on two named vectors configs

Support
1

I have objects with a text title and kicker properties. I will use my own bge-m3 vectorizer. What is the difference between these two styles of using them in the retrieval phase?

vectorizer_config=[
            wvc.config.Configure.NamedVectors.none(source_properties=["title","kicker"],name="title_bge-m3")]

or just do a title+’ '+kicker and vectorize this single field?

Thanks

2

Ciao @rjalex !

You mean passing a list of NamedVectors to vectrorizer_config instead of the usual wvc.config.Configure.Vectorizer?

If that’s the case, you could get the same result of wvc.config.Configure.Vectorizer if you name your NamedVector as “default”.

NamedVectors is more flexible/easier to configure, IMO, as you can specify at the namedvector definition the properties that will be used, instead of defining at the property.

So, at the end, you can define that single named vector. Now that you have named it as title_bge_m3 you will need to specify it while retrieving (target_vector) and when inserting the object with the vector, you will need to pass it as a dict:

collection.data.insert(
    {"text": "This is a test"},
    vector={
        "title_bge-m3": [1,2,3,4]
    }
)
1 Like
3

Muiro obrigado @DudaNogueira

1 Like