How to set a range on vector searches?

[kpzen]

I would like to set a range while doing vector queries, so that instead of fetching the top n results I can specify that I want the 1st, 2nd, 3,rd result and so on.

This seems to be a supported function in several of the integrated vector DBs, see for example Weaviate and Supabase. How do I modify my queries using Langchain to make use of these functions? Is it supported?

[dosubot[bot]]

🤖

Hello,

Yes, LangChain does support range queries in vector searches similar to Weaviate and Supabase. You can use the boolean_filter or efficient_filter parameters in the similarity_search method of the OpenSearchVectorSearch class to set a range for the results. These parameters accept a dictionary that specifies the path to the field, the operator, and the value to filter by.

Here's an example of how you can set a range for the 'page' metadata from 1 to 3:

output = opensearch_vector_search.similarity_search(
    "foo",
    k=2,
    boolean_filter=[
        {"path": ["page"], "operator": "GreaterThanEqual", "valueNumber": 1},
        {"path": ["page"], "operator": "LessThanEqual", "valueNumber": 3},
    ],
)

In this code, boolean_filter is set to a list of two dictionaries. The first dictionary filters the results where the 'page' metadata is greater than or equal to 1, and the second dictionary filters the results where the 'page' metadata is less than or equal to 3.

You can find more details in the source code here:

• libs/langchain/tests/integration_tests/vectorstores/test_weaviate.py
• libs/langchain/langchain/vectorstores/opensearch_vector_search.py

I hope this helps! Let me know if you have any other questions.

Sources

• libs/langchain/tests/integration_tests/vectorstores/test_weaviate.py
• libs/langchain/langchain/vectorstores/opensearch_vector_search.py
• libs/langchain/tests/integration_tests/retrievers/test_weaviate_hybrid_search.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.