rag_node.py

"""
Module for parsing the HTML node
"""

from typing import List
from langchain.docstore.document import Document
from langchain.retrievers import ContextualCompressionRetriever
from langchain.retrievers.document_compressors import EmbeddingsFilter, DocumentCompressorPipeline
from langchain_community.document_transformers import EmbeddingsRedundantFilter
from langchain_community.vectorstores import FAISS
from langchain_openai import OpenAIEmbeddings, AzureOpenAIEmbeddings
from ..models import OpenAI, Ollama, AzureOpenAI
from langchain_community.embeddings import OllamaEmbeddings
from .base_node import BaseNode


class RAGNode(BaseNode):
    """
    A node responsible for compressing the input tokens and storing the document
    in a vector database for retrieval.

    It allows scraping of big documents without exceeding the token limit of the language model.

    Attributes:
        node_name (str): The unique identifier name for the node, defaulting to "ParseHTMLNode".
        node_type (str): The type of the node, set to "node" indicating a standard operational node.

    Args:
        node_name (str, optional): The unique identifier name for the node. 
        Defaults to "ParseHTMLNode".

    Methods:
        execute(state): Parses the HTML document contained within the state using 
        the specified tags, if provided, and updates the state with the parsed content.
    """

    def __init__(self, input: str, output: List[str], node_config: dict, node_name: str = "RAG"):
        """
        Initializes the ParseHTMLNode with a node name.
        """
        super().__init__(node_name, "node", input, output, 2, node_config)
        self.llm_model = node_config["llm"]
        self.embedder_model = node_config.get("embedder_model", None)

    def execute(self, state):
        """
        Executes the node's logic to implement RAG (Retrieval-Augmented Generation) 
        The method updates the state with relevant chunks of the document.

        Args:
            state (dict): The state containing the 'document' key with the HTML content

        Returns:
            dict: The updated state containing the 'relevant_chunks' key with the relevant chunks.

        Raises:
            KeyError: If 'document' is not found in the state, indicating that the necessary 
                      information for parsing is missing.
        """

        print(f"--- Executing {self.node_name} Node ---")

        # Interpret input keys based on the provided input expression
        input_keys = self.get_input_keys(state)

        # Fetching data from the state based on the input keys
        input_data = [state[key] for key in input_keys]

        user_prompt = input_data[0]
        doc = input_data[1]

        chunked_docs = []

        for i, chunk in enumerate(doc):
            doc = Document(
                page_content=chunk,
                metadata={
                    "chunk": i + 1,
                },
            )
            chunked_docs.append(doc)

        print("--- (updated chunks metadata) ---")

        # check if embedder_model is provided, if not use llm_model
        embedding_model = self.embedder_model if self.embedder_model else self.llm_model

        if isinstance(embedding_model, OpenAI):
            embeddings = OpenAIEmbeddings(
                api_key=embedding_model.openai_api_key)
        elif isinstance(embedding_model, AzureOpenAI):
            embeddings = AzureOpenAIEmbeddings()
        elif isinstance(embedding_model, Ollama):
            embeddings = OllamaEmbeddings()
        else:
            raise ValueError("Embedding Model missing or not supported")

        retriever = FAISS.from_documents(
            chunked_docs, embeddings).as_retriever()

        redundant_filter = EmbeddingsRedundantFilter(embeddings=embeddings)
        # similarity_threshold could be set, now k=20
        relevant_filter = EmbeddingsFilter(embeddings=embeddings)
        pipeline_compressor = DocumentCompressorPipeline(
            transformers=[redundant_filter, relevant_filter]
        )
        # redundant + relevant filter compressor
        compression_retriever = ContextualCompressionRetriever(
            base_compressor=pipeline_compressor, base_retriever=retriever
        )

        # relevant filter compressor only
        # compression_retriever = ContextualCompressionRetriever(
        #     base_compressor=relevant_filter, base_retriever=retriever
        # )

        compressed_docs = compression_retriever.get_relevant_documents(
            user_prompt)

        print("--- (tokens compressed and vector stored) ---")

        state.update({self.output[0]: compressed_docs})
        return state