docs/integrations/vectorstores/mongodb_atlas/

[giscus[bot]]

docs/integrations/vectorstores/mongodb_atlas/

This notebook covers how to MongoDB Atlas vector search in LangChain, using the langchain-mongodb package.

https://python.langchain.com/docs/integrations/vectorstores/mongodb_atlas/

[BCazz]

It looks like the links to the Langchain/MongoDB API is non-existent. Is this something that is being worked on or is it no longer supported?

[BCazz]

It looks like the API is back up which is very helpful. Is there anywhere to see all of the parameters for vector_store.as_retreiver() method? I am curious if I can convert the vector store to a retriever but keep the ability the pre-filter the vector search as shown in the Pre-filtering with Similarity Search example.

[jondoescoding]

For me whenever I try to query the mongodb I get back no documents at all.

[BCazz]

Try these for reference:

Makes use of LangChain:
CLIENT = MongoClient(MONGODB_URI, server_api=ServerApi('1'))
DB = CLIENT.CAFGPT
CONTRACT_CHUNKS_COLLECTION = DB['ContractChunks']
ATLAS_VECTOR_SEARCH_INDEX_NAME = "vector_index"
ATLAS_VECTOR_FIELD = "vector" # the field that contains the vector
ATLAS_EMBEDDED_FIELD = "content" # the field that is actually embedded
RELEVANCE_SCORE = "cosine"
EMBEDDINGS = OpenAIEmbeddings(
openai_api_key=OPENAI_API_KEY,
model="text-embedding-3-large"
)

vector_store = MongoDBAtlasVectorSearch(
collection=CONTRACT_CHUNKS_COLLECTION,
embedding=EMBEDDINGS,
index_name=ATLAS_VECTOR_SEARCH_INDEX_NAME,
text_key=ATLAS_EMBEDDED_FIELD,
embedding_key=ATLAS_VECTOR_FIELD,
relevance_score_fn=RELEVANCE_SCORE,
)

async def vector_search(query,document_id):
"""function to run a vector search using the langchain / mongodb vector_store
*cannot use the vector store as a retriever and prefilter the data
Useful Links:
https://langchain-mongodb.readthedocs.io/en/latest/langchain_mongodb/vectorstores/langchain_mongodb.vectorstores.MongoDBAtlasVectorSearch.html
https://python.langchain.com/docs/integrations/vectorstores/mongodb_atlas/#query-directly"""

docs = await vector_store.asimilarity_search(
    query=query,
    k=2,
    pre_filter={'document_id': {'$eq': document_id}}
)

results = []
for doc in docs:
    result = doc.page_content
    results.append(result)

return results

if name == "main":
query = "By which laws shall this agreement be governed by and interpreted in accordance with? Are there any provisions related to the dispute resolution?"
document_id = "9a6ddb6d-4363-4271-9678-96740cbc6e04"

search_result = asyncio.run( vector_search(query,document_id) )

for doc in search_result:
    print(doc)
    print("\n\n")

Makes more of MongoDB:
CLIENT = AsyncMongoClient(MONGODB_URI, server_api=ServerApi('1'))
DB = CLIENT.CAFGPT
CONTRACT_CHUNKS_COLLECTION = DB['ContractChunks']
ATLAS_VECTOR_SEARCH_INDEX_NAME = "vector_index"
ATLAS_VECTOR_PATH = "vector"
EMBEDDINGS = OpenAIEmbeddings(
openai_api_key=OPENAI_API_KEY,
model="text-embedding-3-large"
)

def embed_query(query):
query_vector = openai.embeddings.create(
input=query,
model="text-embedding-3-large"
)
query_vector = query_vector.data[0].embedding
return query_vector

async def vector_search(query,document_id):
"""Call to run a vector similarity search on a MongoDB collection of legal passages within the contract at hand.
Args: query - the value that the similarity search will be run on
Useful Link --> https://www.mongodb.com/docs/atlas/atlas-vector-search/vector-search-stage/"""

# embed query
query_vector = embed_query(query)

# define pipeline
pipeline = [
    {
        '$vectorSearch': {
            'index': ATLAS_VECTOR_SEARCH_INDEX_NAME,
            'path': ATLAS_VECTOR_PATH,
            'filter': {'document_id': {'$eq': document_id}},
            'queryVector': query_vector,
            'numCandidates': 1000,
            'limit': 2
        }
    }, 
    {
        '$project': {
            '_id': 0,
            'content': 1
        }
    }
]

# run pipeline and append the docs to a list
cursor = await CONTRACT_CHUNKS_COLLECTION.aggregate(pipeline)

result = []
async for doc in cursor:
    result.append(doc)

return result 

if name == "main":
query = "By which laws shall this agreement be governed by and interpreted in accordance with? Are there any provisions related to the dispute resolution?"
document_id = "9a6ddb6d-4363-4271-9678-96740cbc6e04"

search_result = asyncio.run( vector_search(query,document_id) )

for doc in search_result:
    print(doc)
    print("\n\n")