ConversationalRetrievalChain doesn't return score with sources

[zigax1]

System Info

LangChain v0.0.171
ChromaDB v0.3.22
Python v3.10.11

Who can help?

No response

Information

• The official example notebooks/scripts
• My own modified scripts

Related Components

• LLMs/Chat Models
• Embedding Models
• Prompts / Prompt Templates / Prompt Selectors
• Output Parsers
• Document Loaders
• Vector Stores / Retrievers
• Memory
• Agents / Agent Executors
• Tools / Toolkits
• Chains
• Callbacks/Tracing
• Async

Reproduction

This is my code:

 def askQuestion(self, collection_id, question):
        collection_name = "collection-" + str(collection_id)
        self.llm = ChatOpenAI(model_name=self.model_name, temperature=self.temperature, openai_api_key=os.environ.get('OPENAI_API_KEY'), streaming=True, callback_manager=CallbackManager([StreamingStdOutCallbackHandler()]))
        self.memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True,  output_key='answer')
        
        chroma_Vectorstore = Chroma(collection_name=collection_name, embedding_function=self.embeddingsOpenAi, client=self.chroma_client)

        self.chain = ConversationalRetrievalChain.from_llm(self.llm, chroma_Vectorstore.as_retriever(),
                                                             return_source_documents=True,verbose=VERBOSE, 
                                                             memory=self.memory)

    
        result = self.chain({"question": question})
        print(result)

        res_dict = {
            "answer": result["answer"],
        }

        res_dict["source_documents"] = []

        for source in result["source_documents"]:
            res_dict["source_documents"].append({
                "page_content": source.page_content,
                "metadata":  source.metadata
            })

        return res_dict

Expected behavior

When I print the result directly after result = self.chain({"question": question}), I get displayed sources, metadata, kwargs, question, chat_history.

I see here: https://github.com/hwchase17/langchain/blob/0c3de0a0b32fadb8caf3e6d803287229409f9da9/langchain/vectorstores/chroma.py#L165 and in line 182 in the official source code, that the similarity_search_with_score() is being called by default.

How can I also display the score than?

[AvikantSrivastava]

Hey @zigax1 if you see the return line of the similarity_search function, only doc is being returned

 return [doc for doc, _ in docs_and_scores] 

https://github.com/hwchase17/langchain/blob/0c3de0a0b32fadb8caf3e6d803287229409f9da9/langchain/vectorstores/chroma.py#L183

Try using the similarity_search_with_score function instead

[zigax1]

@AvikantSrivastava Thank you for the answer, I really appreciate it.
Can you please be more specific, where should I use the similarity_search_with_score? Because I am not using even similarity_search.

I am using ConversationalRetrievalChain which has it integrated. The code snippet is in above.

Thank you

[AvikantSrivastava]

I checked the implementation of VectorStoreRetriever ,

The chain internally callsVectorStoreRetriever get_relevant_documents, and the default value for search_type is 'similarity' that produces no scores

[AvikantSrivastava]

@hwchase17
@agola11
@vowelparrot

Should we add a flag to ConversationalRetrievalChain like similarity = True to get similarity scores from the chain?
I can raise a PR, can you assign me this issue?

[zigax1]

That would be awesome

[vowelparrot]

Thanks for raising this issue!

@zigax1 what do you wish to do with the scores once they're available?

[zigax1]

@vowelparrot, I am using ConversationalRetrievalChain to chat over multiple files (some of them are PDF, some docx, txt,..). So for me is very important to also return the source, so I easily know from which file the answer came.
This however works great if I ask the model question, which is related to the content from some of the documents. If question is very different, like 'What is the distance from moon to mars?' or 'Write me a song about .....' the model still gives me source, but that source is the first embedding in database. This technically is not 'wrong', because this still is source but the distance from vector is very very large.

Here score comes in to make wonders. I could easily than filter the sources before displaying them, if the score is low.

I hope this will be a good enough answer for you to raise the PR for @AvikantSrivastava to integrate this funcionality.

Wish all of you a nice day.

[KeshavSingh29]

@vowelparrot Something similar to the issue here.
The following function is currently not implemented. This makes it difficult to use vectorstore as retriever when we want to return docs and relevance scores.
Any plans to implement it ?

langchain/vectorstores/base.py

    def _similarity_search_with_relevance_scores(
        self,
        query: str,
        k: int = 4,
        **kwargs: Any,
    ) -> List[Tuple[Document, float]]:
        """Return docs and relevance scores, normalized on a scale from 0 to 1.

        0 is dissimilar, 1 is most similar.
        """
        raise NotImplementedError
```

[vowelparrot]

@KeshavSingh29 great question - the short answer is we're working with the maintainers of the vector stores towards that goal.

The longer answer is that each of the vector stores use different distance or similarity functions to compute scores (that also frequently are sensitive to the embeddings you're using). Sometimes it's a distance metric (lower number means more similar), while (less often) it's a similarity metric (higher number is more similar), and the results are usually unbounded.

We could add a default implementation that tries to capture the most common use cases but we have refrained from trying to put a 'one size fits all' normalization function in the base class to avoid providing erroneous results

[zigax1]

Thanks for raising this issue!

@zigax1 what do you wish to do with the scores once they're available?

@vowelparrot What are your thoughts about my answer? Will there be incoming change for ConversationalRetrievalChain to also return score with sources?

Thank you for the answer.

[zigax1]

@AvikantSrivastava do you have any new information about opening a PR? Is solving this in process or not yet?

Thanks

[jezza7770]

Probably a dumb question, but what version of which package do I need to have for this to work? I've just done a pip install yesterday and it's still not returning scores.

[paluchasz]

Also doesn't seem to work for me still

[khaledalarja]

This issue should be reopened, the issue is still not solved, the PR #5655 is not related to the behavior of ConversationalRetrievalChain, it's only specific for Chroma, but here it's the problem, scores are neglected with the _

[DmitryKatson]

No chance we can return source documents with scores now, with ConversationalRetrievalChain.from_llm?

[prasoons075]

is this resolved? I am trying to get a score when using ConversationalRetrivalChain with the FAISS vector store.

[naarkhoo]

looks this still only return the doc but not the score;

langchain/langchain/vectorstores/faiss.py

Line 206 in 0c3de0a

return docs

[baswenneker]

Here's my hacky solution. It's a retriever that saves the score as part of the documents metadata. I just took the original _get_relevant_documents and rewrote the part @khaledalarja pointed us at in his comment.

class MyVectorStoreRetriever(VectorStoreRetriever):
    # See https://github.com/langchain-ai/langchain/blob/61dd92f8215daef3d9cf1734b0d1f8c70c1571c3/libs/langchain/langchain/vectorstores/base.py#L500
    def _get_relevant_documents(
        self, query: str, *, run_manager: CallbackManagerForRetrieverRun
    ) -> List[Document]:
        docs_and_similarities = (
            self.vectorstore.similarity_search_with_relevance_scores(
                query, **self.search_kwargs
            )
        )

        # Make the score part of the document metadata
        for doc, similarity in docs_and_similarities:
            doc.metadata["score"] = similarity

        docs = [doc for doc, _ in docs_and_similarities]
        return docs

Instead of doing vectordb.as_retriever I instantiate the retriever by doing:

retriever = MyVectorStoreRetriever(
   vectorstore=vectordb,
   search_type="similarity_score_threshold",
   search_kwargs={"score_threshold": similarity_threshold, "k": 3},
)

Good luck :)

[matardy]

Here's my hacky solution. It's a retriever that saves the score as part of the documents metadata. I just took the original _get_relevant_documents and rewrote the part @khaledalarja pointed us at in his comment.

class MyVectorStoreRetriever(VectorStoreRetriever):
    # See https://github.com/langchain-ai/langchain/blob/61dd92f8215daef3d9cf1734b0d1f8c70c1571c3/libs/langchain/langchain/vectorstores/base.py#L500
    def _get_relevant_documents(
        self, query: str, *, run_manager: CallbackManagerForRetrieverRun
    ) -> List[Document]:
        docs_and_similarities = (
            self.vectorstore.similarity_search_with_relevance_scores(
                query, **self.search_kwargs
            )
        )

        # Make the score part of the document metadata
        for doc, similarity in docs_and_similarities:
            doc.metadata["score"] = similarity

        docs = [doc for doc, _ in docs_and_similarities]
        return docs

Instead of doing vectordb.as_retriever I instantiate the retriever by doing:

retriever = MyVectorStoreRetriever(
   vectorstore=vectordb,
   search_type="similarity_score_threshold",
   search_kwargs={"score_threshold": similarity_threshold, "k": 3},
)

Good luck :)

Thanks, this works totally fine!

[Benvorth]

Took me a bit to find the correct imports :

from langchain.schema.vectorstore import VectorStoreRetriever
from langchain.callbacks.manager import CallbackManagerForRetrieverRun
from langchain.schema.document import Document
from typing import List
...

[glaucoleme]

Here's my hacky solution. It's a retriever that saves the score as part of the documents metadata. I just took the original _get_relevant_documents and rewrote the part @khaledalarja pointed us at in his comment.

class MyVectorStoreRetriever(VectorStoreRetriever):
    # See https://github.com/langchain-ai/langchain/blob/61dd92f8215daef3d9cf1734b0d1f8c70c1571c3/libs/langchain/langchain/vectorstores/base.py#L500
    def _get_relevant_documents(
        self, query: str, *, run_manager: CallbackManagerForRetrieverRun
    ) -> List[Document]:
        docs_and_similarities = (
            self.vectorstore.similarity_search_with_relevance_scores(
                query, **self.search_kwargs
            )
        )

        # Make the score part of the document metadata
        for doc, similarity in docs_and_similarities:
            doc.metadata["score"] = similarity

        docs = [doc for doc, _ in docs_and_similarities]
        return docs

Instead of doing vectordb.as_retriever I instantiate the retriever by doing:

retriever = MyVectorStoreRetriever(
   vectorstore=vectordb,
   search_type="similarity_score_threshold",
   search_kwargs={"score_threshold": similarity_threshold, "k": 3},
)

Good luck :)

Also worked for me. This could be considered as a final solution to this PR.