DEVOCEAN

import os
from langchain_openai import AzureChatOpenAI
from langchain_openai import AzureOpenAIEmbeddings

embeddings = AzureOpenAIEmbeddings(
    model="text-embedding-ada-002",
    openai_api_version="2024-02-15-preview",
    azure_endpoint=os.getenv("AZURE_ENDPOINT"),
    api_key=os.getenv("AZURE_OPENAI_KEY"),
)

llm = AzureChatOpenAI(
    model="gpt-4-32k",
    azure_endpoint = os.getenv("AZURE_ENDPOINT"),
    api_key=os.getenv("AZURE_OPENAI_KEY"),  
    api_version="2024-02-15-preview"
)

from langchain_community.document_loaders import WebBaseLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import Chroma

loader = WebBaseLoader(
    "https://blog.langchain.dev/langchain-v0-1-0/"
)
documents = loader.load()

text_splitter = RecursiveCharacterTextSplitter(
    chunk_size = 1250,
    chunk_overlap = 100,
    length_function = len,
    is_separator_regex = False
)
split_docs = text_splitter.split_documents(documents)
vectorstore = Chroma.from_documents(
    documents=split_docs, 
    embedding=embeddings, 
    persist_directory="./Vectorstore/chromadb"
)

from operator import itemgetter

from langchain import hub
from langchain_core.runnables import RunnablePassthrough
from langchain_core.output_parsers import StrOutputParser

retriever = vectorstore.as_retriever()
prompt = hub.pull("rlm/rag-prompt")

def format_docs(args):
    docs = args["source_docs"]
    return "\n\n".join(doc.page_content for doc in docs)

rag_chain = (
    RunnablePassthrough.assign(
        question=itemgetter("question")
    )
    | {
        "source_docs" : itemgetter("question") | retriever,
        "question": itemgetter("question"),
    }
    | {
        "source_docs": itemgetter("source_docs"), 
        "answer": {
            "context": format_docs,
            "question": itemgetter("question"),        
        } | prompt | llm | StrOutputParser()
    }
)

rag_chain.invoke({"question" : "What are the major changes in v 0.1.0?"})

{'source_docs': [Document(page_content='versioning policy for a little over a month now.langchain itself, however, still remained on 0.0.x versions. ... See more towards the end of this post on our plans for that.While re-architecting the', metadata={'language': 'en', 'source': 'https://blog.langchain.dev/langchain-v0-1-0/', 'title': 'LangChain v0.1.0'}),
  ... ],
 'answer': 'The major changes in v 0.1.0 include a new versioning standard where breaking changes to the public API result in a minor version bump and bug fixes or new features result in a patch version bump. Additionally, the v 0.1.0 release of langchain is the first stable version, fully backwards compatible, available in both Python and JavaScript with improved focus and documentation. Architectural changes were made such as separating langchain-core and partner packages from langchain and old, unused functionality was removed.'}

from ragas.testset.generator import TestsetGenerator
from ragas.testset.evolutions import simple, reasoning, multi_context

generator = TestsetGenerator.from_langchain(
    generator_llm=llm,
    critic_llm=llm,
    embeddings=embeddings
)
testset = generator.generate_with_langchain_docs(split_docs, test_size=10, distributions={simple: 0.5, reasoning: 0.25, multi_context: 0.25})

test_df = testset.to_pandas()
test_questions = test_df["question"].values.tolist()
test_groundtruths = test_df["ground_truth"].values.tolist()
test_df.head()

from datasets import Dataset

answers = []
contexts = []

for question in test_questions:
  response = rag_chain.invoke({"question": question})
  answers.append(response['answer'])
  contexts.append([context.page_content for context in response['source_docs']])

response_dataset = Dataset.from_dict({
    "question" : test_questions,
    "answer" : answers,
    "contexts" : contexts,
    "ground_truth" : test_groundtruths
})

from ragas import evaluate
from ragas.metrics import (
    faithfulness,
    answer_relevancy,
    answer_correctness,
    context_recall,
    context_precision,
)

metrics = [
    faithfulness,
    answer_relevancy,
    context_recall,
    context_precision,
    answer_correctness,
]

naive_results = evaluate(response_dataset, metrics, raise_exceptions=False)

{
 'faithfulness': 1.0000, 
 'answer_relevancy': 0.9347, 
 'context_recall': 1.0000, 
 'context_precision': 0.8750, 
 'answer_correctness': 0.6409
}