New examples added to docs

[hari0205]

• Added a new example on how to RAG using metadata on retrievers
• Some linting done.

[hari0205]

@hsm207 is this good enough for a PR? I noticed that as_retriever docstring on langchain_core vector stores is a bit misleading and incorrect.

[hsm207]

@hari0205 could you please clarify what part of the as_retriever is a bit misleading and incorrect? Since langchain-weaviate does not override that method, I'm inclined to think that the clarification on as_retriever should go to the main langchain repo

cc @efriis

[hari0205]

@hari0205 could you please clarify what part of the as_retriever is a bit misleading and incorrect? Since langchain-weaviate does not override that method, I'm inclined to think that the clarification on as_retriever should go to the main langchain repo

cc @efriis

The as_retriever docstring mentions that it takes filter as one of the keyword arguments:

 # Use a filter to only retrieve documents from a specific paper
            docsearch.as_retriever(
                search_kwargs={'filter': {'paper_title':'GPT-4 Technical Report'}}
            )

So, I tried an example

import weaviate as wev
from langchain_weaviate import WeaviateVectorStore

client = wev.connect_to_local(port=8030)
db = WeaviateVectorStore(
    client, "myindex", "mytext", embedding=OpenAIEmbeddings(), by_text=False
)

retriver = db.as_retriever(search_kwargs={"filters":{"userid":userid}})
llm = ChatOpenAI(model="gpt-4-turbo-preview",temperature=0)
    
qa_chain = RetrievalQA.from_chain_type(llm,retriever=retriver)
res=qa_chain.invoke({"query":query})
print(res)

I got a type error:

 File "E:\work\beyondcc\resume_parse\langchain\Lib\site-packages\langchain\chains\base.py", line 162, in invoke
    raise e
  File "E:\work\beyondcc\resume_parse\langchain\Lib\site-packages\langchain\chains\base.py", line 156, in invoke
    self._call(inputs, run_manager=run_manager)
  File "E:\work\beyondcc\resume_parse\langchain\Lib\site-packages\langchain\chains\retrieval_qa\base.py", line 141, in _call
    docs = self._get_docs(question, run_manager=_run_manager)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\work\beyondcc\resume_parse\langchain\Lib\site-packages\langchain\chains\retrieval_qa\base.py", line 221, in _get_docs
    return self.retriever.get_relevant_documents(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\work\beyondcc\resume_parse\langchain\Lib\site-packages\langchain_core\retrievers.py", line 245, in get_relevant_documents
    raise e
  File "E:\work\beyondcc\resume_parse\langchain\Lib\site-packages\langchain_core\retrievers.py", line 238, in get_relevant_documents
    result = self._get_relevant_documents(
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\work\beyondcc\resume_parse\langchain\Lib\site-packages\langchain_core\vectorstores.py", line 696, in _get_relevant_documents
    docs = self.vectorstore.similarity_search(query, **self.search_kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\work\beyondcc\resume_parse\langchain\Lib\site-packages\langchain_weaviate\vectorstores.py", line 288, in similarity_search
    result = self._perform_search(query, k, **kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\work\beyondcc\resume_parse\langchain\Lib\site-packages\langchain_weaviate\vectorstores.py", line 241, in _perform_search
    result = collection.query.hybrid(
             ^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: _HybridQuery.hybrid() got an unexpected keyword argument 'filter'

Upon closer inspection on class WeaviateVectorStore(VectorStore): in langchain_weviate, I was able to figure out that the hybrid method takes filters instead of filter.

I changed the code to

retriver = db.as_retriever(search_kwargs={"filters":{"userid":userid}})

Now, I am greeted with another error

  File "E:\work\beyondcc\resume_parse\langchain\Lib\site-packages\weaviate\collections\queries\hybrid\query.py", line 84, in hybrid
    res = self._query.hybrid(
          ^^^^^^^^^^^^^^^^^^^
  File "E:\work\beyondcc\resume_parse\langchain\Lib\site-packages\weaviate\collections\grpc\query.py", line 194, in hybrid
    request = self.__create_request(
              ^^^^^^^^^^^^^^^^^^^^^^
  File "E:\work\beyondcc\resume_parse\langchain\Lib\site-packages\weaviate\collections\grpc\query.py", line 521, in __create_request
    _validate_input(
  File "E:\work\beyondcc\resume_parse\langchain\Lib\site-packages\weaviate\validator.py", line 24, in _validate_input
    raise WeaviateInvalidInputError(
weaviate.exceptions.WeaviateInvalidInputError: Invalid input provided: Argument 'filters' must be one of: [<class 'weaviate.collections.classes.filters._Filters'>, None], but got <class 'dict'>.

The only way I was able to get it working was to modify the code to

search_filter=wev.classes.query.Filter.by_property("userid").equal(userid)
retriver = db.as_retriever(search_kwargs={"filters":search_filter})
llm = ChatOpenAI(model="gpt-4-turbo-preview",temperature=0)
    
qa_chain = RetrievalQA.from_chain_type(llm,retriever=retriver)
res=qa_chain.invoke({"query":query})
print(res)

My assumption is that the implementation of metadata filter is different for different vector stores. Since the keyword arguments are passed as they are, we encounter this error. So, I opened this PR to provide some examples and avoid confusion.

PS: package versions

langchain==0.1.6
langchain-community==0.0.19
langchain-core==0.1.38
langchain-openai==0.0.6
langchain-text-splitters==0.0.1
langchain-weaviate==0.0.1rc5