Why does my retrieval config doesn't work for topK?

[lingvisa]

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

from langchain.vectorstores import Chroma
from chromadb_loader import embeddings
from langchain_community.vectorstores import Chroma
import chromadb
import pickle
from langchain_core.runnables import ConfigurableField
from langchain.retrievers import EnsembleRetriever

# Specify the directory where the database is persisted
persist_directory = './chroma_data'

persist_client = chromadb.PersistentClient(path=persist_directory)
vectordb = Chroma(client=persist_client, collection_name='test', embedding_function=embeddings)
available_collections = [c.name for c in persist_client.list_collections()]
# Load the vector database from the persisted directory

# Perform a similarity search to check that the database is loaded correctly
question = "AI"
docs = vectordb.similarity_search_with_score(question, k=1)


with open('./bm25_retriever.pkl', 'rb') as f:
    keyword_retriever = pickle.load(f)

retriever = vectordb.as_retriever(search_kwargs={"k": 2}).configurable_fields(
    search_kwargs=ConfigurableField(
        id="search_kwargs_chroma",
        name="Search Kwargs",
        description="The search kwargs to use",
    )
)
ensemble_retriever = EnsembleRetriever(retrievers=[retriever,keyword_retriever],weights=[0.9, 0.1], return_source_documents=True)

config = {"configurable": {"search_kwargs_chroma": {"k": 1}}}
docs2 = ensemble_retriever.invoke(question, config=config)
# Print the results
for doc, score in docs:
    print(doc,score)
print('Ensemble results: ')
for doc in docs2:
    print(doc)

Description

Specifically, the 'k'=1 configuration doesn't work for me.

docs2 = ensemble_retriever.invoke(question, config=config). it returns 5 results.

System Info

langchain==0.2.11
langchain-aws==0.1.6
langchain-community==0.2.10
langchain-core==0.2.23
langchain-google-genai==1.0.7
langchain-openai==0.1.17
langchain-text-splitters==0.2.2
langchain-weaviate==0.0.2

[dosubot[bot]]

Hey @lingvisa! 👋 I'm here to help you with bug fixes, answer questions, and guide you to become a contributor. Let's tackle this together!

The issue with your retrieval config not working for topK might be due to the validation of the top_k parameter. The top_k parameter must be a non-negative integer. If you set top_k to a negative value, it will raise a ValueError.

Here is the relevant code snippet that validates the top_k parameter:

@validator("top_k")
def validate_top_k(cls, value: int) -> int:
    if value < 0:
        raise ValueError(f"top_k ({value}) cannot be negative.")
    return value

Additionally, ensure that the k value you are setting is appropriate and non-negative. In the context of the EnsembleRetriever, the config parameter is used to set the search_kwargs for the individual retrievers at runtime. The EnsembleRetriever correctly propagates this configuration, resulting in the retrievers returning the specified number of results.

Here is an example of how the config parameter is used:

from langchain_core.runnables import ConfigurableField

faiss_retriever = faiss_vectorstore.as_retriever(
    search_kwargs={"k": 2}
).configurable_fields(
    search_kwargs=ConfigurableField(
        id="search_kwargs_faiss",
        name="Search Kwargs",
        description="The search kwargs to use",
    )
)

ensemble_retriever = EnsembleRetriever(
    retrievers=[bm25_retriever, faiss_retriever], weights=[0.5, 0.5]
)

config = {"configurable": {"search_kwargs_faiss": {"k": 1}}}
docs = ensemble_retriever.invoke("apples", config=config)
docs

In your case, ensure that the config parameter is correctly set and propagated to the individual retrievers. The EnsembleRetriever uses the patch_config function to ensure that the configuration is correctly applied to each retriever during the invocation process ^[1][2].

If the issue persists, double-check the versions of the LangChain modules you are using, as there might be compatibility issues between different versions ^[3][4].

To continue talking to Dosu, mention @dosu.