Choma similarity_search_with_score (and similar methods) don't populate the document.id property of the returned documents. #26860
Comments
|
Until there is a fix I am patching my code with this helper method that has my fix in it: ####################################################################################################
# This is PATCHED code from the original langchain_chroma module
####################################################################################################
# See bug: https://github.com/langchain-ai/langchain/issues/26860
def similarity_search_with_score_with_id(
vector_db: Chroma,
query: str,
k: int = DEFAULT_K,
filter: Optional[dict[str, str]] = None,
where_document: Optional[dict[str, str]] = None,
**kwargs,
) -> list[tuple[Document, float]]:
if vector_db._embedding_function is None:
results = vector_db._collection.query(
query_texts=[query],
n_results=k,
where=filter, # type: ignore
where_document=where_document, # type: ignore
**kwargs,
)
else:
query_embedding = vector_db._embedding_function.embed_query(query)
results = vector_db._collection.query(
query_embeddings=[query_embedding], # type: ignore
n_results=k,
where=filter, # type: ignore
where_document=where_document, # type: ignore
**kwargs,
)
return _results_to_docs_and_scores(results)
def _results_to_docs_and_scores(results) -> list[tuple[Document, float]]:
return [
# TODO: Chroma can do batch querying,
# we shouldn't hard code to the 1st result
(Document(id =result[0], page_content=result[1], metadata=result[2] or {}), result[3])
for result in zip(
results["ids"][0],
results["documents"][0],
results["metadatas"][0],
results["distances"][0],
)
]
####################################################################################################
# End of PATCHED code
####################################################################################################Incase anyone finds it useful |
|
Not all vectorstores currently support this. Someone can make a PR to add support for returning the ID as part of the document. The ID in documents was added only a few months ago, and it still considered optional |
|
Hi @eyurtsev, |
|
@KrisTC ha sorry i didn't mean to imply that. This was for anyone in the community who wants to tackle this. I should read more carefully what I write :) |
|
@eyurtsev We are looking to contribute to this issue. Since @KrisTC already provided a fix for Chroma, we can potentially write test cases for it and also investigate the remaining vectorstores and add support for the ones that are missing this functionality. I believe we are targeting only vectorstores with similarity search, would that be correct? If you have more detailed information on which vectorstores need an update, that would be greatly appreciated. |
|
There's a standard test suite in the standard tests package that covers a lot of the relevant edge cases. You can git grep through the codebase to find example usage git grep "standard_test" |
|
Some update, any suggestions are appreciated:
|
|
Hello, @eyurtsev We have the fix and test cases for ChromaDB. Should we create a pull request for Chroma DB first and work on the other vector stores later on other pull request only or should we create a pull request after fixing other vector stores altogether |
|
Individual PRs would be great |
|
Just created a PR #27366 |
**Description:** Returns the document id along with the Vector Search results **Issue:** Fixes #26860 for CouchbaseVectorStore - [x] **Add tests and docs**: If you're adding a new integration, please include 1. a test for the integration, preferably unit tests that do not rely on network access, 2. an example notebook showing its use. It lives in `docs/docs/integrations` directory. - [x] **Lint and test**: Run `make format`, `make lint` and `make test` from the root of the package(s) you've modified. Co-authored-by: Erick Friis <[email protected]>
Checked other resources
Example Code
I want to get the document id from the documents I find with
similarity_search_with_scorebut the id's aren't being set to the documents when they are created.Here is my test code:
Error Message and Stack Trace (if applicable)
Test output:
I would expect the result:
Description
I am using:
I am calling
vector_db.similarity_search_with_score(...)orvector_db.similarity_search(...)and the documents returned never have an id.I can't see a way of getting the documents give me an id. I though about adding a kwargs include ids. But the documentation for the underlying query says it always returns ids. I think the issue is in your method:
from:
langchain/libs/partners/chroma/langchain_chroma/vectorstores.py
Lines 43 to 53 in 51c4393
It doesn't use the id.
This fix should be very easy:
System Info
System Information
Package Information
Optional packages not installed
Other Dependencies
The text was updated successfully, but these errors were encountered: