Why do SelfQueryRetrievers require translators in JS & not in Python?

[FaresKi]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

In Python, a self querying retriever looks something like this:

docs = [
    Document(
        page_content="A bunch of scientists bring back dinosaurs and mayhem breaks loose",
        metadata={"year": 1993, "rating": 7.7, "genre": "science fiction"},
    ),
    Document(
        page_content="Leo DiCaprio gets lost in a dream within a dream within a dream within a ...",
        metadata={"year": 2010, "director": "Christopher Nolan", "rating": 8.2},
    ),
    Document(
        page_content="A psychologist / detective gets lost in a series of dreams within dreams within dreams and Inception reused the idea",
        metadata={"year": 2006, "director": "Satoshi Kon", "rating": 8.6},
    ),
    Document(
        page_content="A bunch of normal-sized women are supremely wholesome and some men pine after them",
        metadata={"year": 2019, "director": "Greta Gerwig", "rating": 8.3},
    ),
    Document(
        page_content="Toys come alive and have a blast doing so",
        metadata={"year": 1995, "genre": "animated"},
    ),
    Document(
        page_content="Three men walk into the Zone, three men walk out of the Zone",
        metadata={
            "year": 1979,
            "rating": 9.9,
            "director": "Andrei Tarkovsky",
            "genre": "science fiction",
        },
    ),
]

vectara = Vectara() # or some other vector store
for doc in docs:
    vectara.add_texts([doc.page_content], doc_metadata=doc.metadata)

metadata_field_info = [
    AttributeInfo(
        name="genre",
        description="The genre of the movie",
        type="string or list[string]",
    ),
    AttributeInfo(
        name="year",
        description="The year the movie was released",
        type="integer",
    ),
    AttributeInfo(
        name="director",
        description="The name of the movie director",
        type="string",
    ),
    AttributeInfo(
        name="rating", description="A 1-10 rating for the movie", type="float"
    ),
]
document_content_description = "Brief summary of a movie"
llm = ChatOpenAI(temperature=0, model="gpt-4o", max_tokens=4069)
retriever = SelfQueryRetriever.from_llm(
    llm, vectara, document_content_description, metadata_field_info, verbose=True
)

In JS, it's a similar process, but there's an additional layer of abstraction because there's a structuredQueryTranslator. Just wondering why.

If I want to use this feature in JS with PGVectors, it isn't clear how.

Description

Trying to understand why the differences between the Python & JS SDKs for SelfQueryingRetrievers.

In my humble opinion both methods should be iso enough to be able to swap in terms of method signature (unless something's happening at the vector store level providing the self querying capabilities).

System Info

langchain@0.3.7
mac
22.10.0
N/A

[dosubot[bot]]

The structuredQueryTranslator in the JavaScript implementation of SelfQueryRetrievers is necessary to translate natural language queries into a structured format that the vector store can understand. This is crucial for enabling complex queries that involve filtering based on metadata attributes, such as genre, year, director, or rating ^[1].

In contrast, the Python implementation might handle this translation internally or differently, possibly due to differences in how the language models or vector stores are integrated and utilized in Python versus JavaScript. The need for a structuredQueryTranslator in JavaScript could be due to architectural differences or design choices in the SDKs, which might not be as explicit in Python.

If you're looking to use this feature with PGVectors in JavaScript, you would need to ensure that the structuredQueryTranslator is compatible with PGVectors, or implement a custom translator if necessary. Unfortunately, specific examples or documentation on using SelfQueryRetrievers with PGVectors in the JavaScript SDK were not found in the available resources.

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other}

[FaresKi]

@dosu
Would the FunctionalTranslator work PGVectors? Or at least the HNSW Libs?
I want to leverage PGVectors & SelfQueryRetrievers?

[dosubot[bot]]

Yes, the FunctionalTranslator works with the HNSW libraries when using SelfQueryRetrievers. There are test cases that demonstrate the use of FunctionalTranslator with SelfQueryRetriever and HNSWLib to retrieve relevant documents based on structured queries ^[1]. However, there is no specific mention of PGVectors in the context of FunctionalTranslator compatibility, so I cannot confirm its compatibility with PGVectors.

LangChainJS does have existing integrations for using PGVectors. The integration is available through the @langchain/community package, and it requires the pg package as a peer dependency. The PGVectorStore class is used for this integration, and it supports adding, deleting, and searching documents with similarity scores ^[2][3].

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other}