Retrieval from a single multimodal vector space which was created using both image and text fields - Is there a search operator for this?

Support
1

Description

    Configure.NamedVectors.multi2vec_palm(
        name="vector_space",                    # the name of the combined vector space
        image_fields=["poster"],                 # use poster property multivec vectorization
        
        text_fields=["title", "overview"],
        
        project_id="semi-random-dev",
        location="us-central1",
        model_id="multimodalembedding@001",
        dimensions=1408,
    ),
]

For a given vecotor space created as above, I am looking for a search operator that can search a combination of the above text fields i.e. I would provide the title, overview and the poster as search parameters and it would need to do a a combination of these exactly as it did during creation of the vector space.

2

hi @Vignesh_Iyer !!

Welcome to our community :hugs:

Right now you can only search one vector space at a time. So you can do a similarity search on those two properties, or also a hybrid search, pointing vector_space as target_vector and even specify query properties for the key word part

However, I am considering here that poster is a third property, not included on the vector_space.

If that’s the case, you cannot query multiple named spaces for now.

But good news: this feature is already in the oven and getting ready to be released :wink:

Let me know if this helps!

THanks!