llama_index/llama-index-integrations/vec...

At a glance

The community member is working on a project using the LlamaIndex package and Pinecone for their index. They encountered a ValueError stating that the vector store only supports exact match filters when they attempted to use filters with operators other than exact matches (e.g., FilterOperator.LTE). The community member investigated the issue and found that the LLM was giving the right translation from the query to the metadata filter, and Pinecone also supports the '<=' operation. However, they were not passing in the storage context properly while creating the index, which was causing the error in an incorrect spot. Another community member suggested that the issue might be due to the community member using the simple vector store instead of the Pinecone vector store.

Useful resources

NNehil

Need help figuring out why Pinecone VectorIndexAutoRetriever is not working.

#### Background
I am working on a project using the LlamaIndex package integrating with Pinecone for my index. The project involves creating a VectorIndexAutoRetriever to set up a retriever similar to the example in https://docs.llamaindex.ai/en/stable/examples/vector_stores/pinecone_auto_retriever/

#### Problem Description
During the implementation, I encountered a ValueError stating that the vector store only supports exact match filters. Th error message is:

Plain Text

ValueError: Vector Store only supports exact match filters. Please use ExactMatchFilter or FilterOperator.EQ instead.

This error occurs when I attempt to use filters with operators other than exact matches (e.g., FilterOperator.LTE).

#### Investigation

Filter Setup: The LLM gives the right translation from query to metadata filter. The query is Who are some founders that are 30 or below

Plain Text

MetadataFilters(filters=[MetadataFilter(key='age', value=30, operator=<FilterOperator.LTE: '<='>)], condition=<FilterCondition.AND: 'and'>)

Pinecone also supports this operation '<=' as described by the code below:

Plain Text

def _transform_pinecone_filter_operator(operator: str) -> str:
    """Translate standard metadata filter operator to Pinecone specific spec."""
    print(f"DEBUG: The incoming operator is {operator} from the code")
    ...
    elif operator == ">=":
        return "$gte"
    elif operator == "<=":
        return "$lte"
    ...

GitHub link

What am I doing wrong? Is only exact match filtering supported? I can help add documentation but currently its pretty confusing to debug further for me.

5 comments

LLogan M

Do you have the full traceback?

NNehil

Yep 1 sec

NNehil

Some more useful logs

Plain Text

Vector store index created and ready for use.
DEBUGGG NEHIL [MetadataInfo(name='name', type='str', description='Name of the founder in the profile. Usually in headers'), MetadataInfo(name='profile_url', type='str', description='The url of the profile which is the url of the page'), MetadataInfo(name='linkedin_url', type='str', description='A linked in profile url attached to the profile. Could be None but very rarely.'), MetadataInfo(name='hobbies', type='list[str]', description='List of hobbies of the founder. ex: Cycling, Reading, Gaming'), MetadataInfo(name='employement_industries', type='list[str]', description='List of industries the founder has worked in. ex: AI, Real Estate, Travel'), MetadataInfo(name='location', type='str', description='Location of the founder. Usually in the format location: city, state, country. ex: location: New York, NY, USA'), MetadataInfo(name='age', type='int', description='Age of the founder. Usually in the format age: x. ex: age: 25')]
DEBUGING str_or_query_bundle in _retrieve Who are some founders that are 30 or below
INFO:httpx:HTTP Request: POST https://api.openai.com/v1/chat/completions "HTTP/1.1 200 OK"
HTTP Request: POST https://api.openai.com/v1/chat/completions "HTTP/1.1 200 OK"
INFO:llama_index.core.indices.vector_store.retrievers.auto_retriever.auto_retriever:Using query str: Profiles of founders from startup school YC platform
Using query str: Profiles of founders from startup school YC platform
INFO:llama_index.core.indices.vector_store.retrievers.auto_retriever.auto_retriever:Using filters: [('age', '<=', 30)]
Using filters: [('age', '<=', 30)]
Using query str: Profiles of founders from startup school YC platform
Using filters: [('age', '<=', 30)]
INFO:llama_index.core.indices.vector_store.retrievers.auto_retriever.auto_retriever:Using top_k: 2
Using top_k: 2
DEBUGING str_or_query_bundle in _retrieve Profiles of founders from startup school YC platform

NNehil

@Logan M I was able to figure it out. I was not passing in storage context properly while creating the index. It was giving an error in an incorrect spot. So there might be a bug in where the error is caught.

LLogan M

Ah yes good catch. Following the traceback, we can see it's using the simple vector store, rather than your pinecone vector store 👍

Add a reply

Find answers from the community

llama_index/llama-index-integrations/vec...

Need help figuring out why Pinecone VectorIndexAutoRetriever is not working.