app.py

import os
import json
import re
import gradio as gr
import pandas as pd
import requests
import random
import urllib.parse
from tempfile import NamedTemporaryFile
from typing import List
from bs4 import BeautifulSoup
from langchain_core.prompts import ChatPromptTemplate
from langchain_community.vectorstores import FAISS
from langchain_community.document_loaders import PyPDFLoader
from langchain_core.output_parsers import StrOutputParser
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.llms import HuggingFaceHub
from langchain_core.runnables import RunnableParallel, RunnablePassthrough
from langchain_core.documents import Document
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
from datetime import datetime
from huggingface_hub.utils import HfHubHTTPError

huggingface_token = os.environ.get("HUGGINGFACE_TOKEN")

# Memory database to store question-answer pairs
memory_database = {}
conversation_history = []

def load_and_split_document_basic(file):
    """Loads and splits the document into pages."""
    loader = PyPDFLoader(file.name)
    data = loader.load_and_split()
    return data

def load_and_split_document_recursive(file: NamedTemporaryFile) -> List[Document]:
    """Loads and splits the document into chunks."""
    loader = PyPDFLoader(file.name)
    pages = loader.load()
    
    text_splitter = RecursiveCharacterTextSplitter(
        chunk_size=1000,
        chunk_overlap=200,
        length_function=len,
    )
    
    chunks = text_splitter.split_documents(pages)
    return chunks

def get_embeddings():
    return HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")

def create_or_update_database(data, embeddings):
    if os.path.exists("faiss_database"):
        db = FAISS.load_local("faiss_database", embeddings, allow_dangerous_deserialization=True)
        db.add_documents(data)
    else:
        db = FAISS.from_documents(data, embeddings)
    db.save_local("faiss_database")

def clear_cache():
    if os.path.exists("faiss_database"):
        os.remove("faiss_database")
        return "Cache cleared successfully."
    else:
        return "No cache to clear."

def get_similarity(text1, text2):
    vectorizer = TfidfVectorizer().fit_transform([text1, text2])
    return cosine_similarity(vectorizer[0:1], vectorizer[1:2])[0][0]

prompt = """
Answer the question based on the following information:

Conversation History:
{history}

Context from documents:
{context}

Current Question: {question}

If the question is referring to the conversation history, use that information to answer.
If the question is not related to the conversation history, use the context from documents to answer.
If you don't have enough information to answer, say so.

Provide a concise and direct answer to the question:
"""

def get_model(temperature, top_p, repetition_penalty):
    return HuggingFaceHub(
        repo_id="mistralai/Mistral-7B-Instruct-v0.3",
        model_kwargs={
            "temperature": temperature,
            "top_p": top_p,
            "repetition_penalty": repetition_penalty,
            "max_length": 1000
        },
        huggingfacehub_api_token=huggingface_token
    )

def generate_chunked_response(model, prompt, max_tokens=200):
    full_response = ""
    total_length = len(prompt.split())  # Approximate token count of prompt
    
    while total_length < 7800:  # Leave some margin
        try:
            chunk = model(prompt + full_response, max_new_tokens=min(200, 7800 - total_length))
            chunk = chunk.strip()
            if not chunk:
                break
            full_response += chunk
            total_length += len(chunk.split())  # Approximate token count
            
            if chunk.endswith((".", "!", "?")):
                break
        except Exception as e:
            print(f"Error generating response: {str(e)}")
            break
    
    return full_response.strip()

def manage_conversation_history(question, answer, history, max_history=5):
    history.append({"question": question, "answer": answer})
    if len(history) > max_history:
        history.pop(0)
    return history

def is_related_to_history(question, history, threshold=0.3):
    if not history:
        return False
    history_text = " ".join([f"{h['question']} {h['answer']}" for h in history])
    similarity = get_similarity(question, history_text)
    return similarity > threshold

def extract_text_from_webpage(html):
    soup = BeautifulSoup(html, 'html.parser')
    for script in soup(["script", "style"]):
        script.extract()  # Remove scripts and styles
    text = soup.get_text()
    lines = (line.strip() for line in text.splitlines())
    chunks = (phrase.strip() for line in lines for phrase in line.split("  "))
    text = '\n'.join(chunk for chunk in chunks if chunk)
    return text

_useragent_list = [
    "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36",
    "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36",
    "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Edge/91.0.864.59 Safari/537.36",
    "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Edge/91.0.864.59 Safari/537.36",
    "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Safari/537.36",
    "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Safari/537.36",
]

def google_search(term, num_results=20, lang="en", timeout=5, safe="active", ssl_verify=None):
    escaped_term = urllib.parse.quote_plus(term)
    start = 0
    all_results = []
    max_chars_per_page = 8000  # Limit the number of characters from each webpage to stay under the token limit

    print(f"Starting Google search for term: '{term}'")

    with requests.Session() as session:
        while start < num_results:
            try:
                user_agent = random.choice(_useragent_list)
                headers = {
                    'User-Agent': user_agent
                }
                resp = session.get(
                    url="https://www.google.com/search",
                    headers=headers,
                    params={
                        "q": term,
                        "num": num_results - start,
                        "hl": lang,
                        "start": start,
                        "safe": safe,
                    },
                    timeout=timeout,
                    verify=ssl_verify,
                )
                resp.raise_for_status()
                print(f"Successfully retrieved search results page (start={start})")
            except requests.exceptions.RequestException as e:
                print(f"Error retrieving search results: {e}")
                break

            soup = BeautifulSoup(resp.text, "html.parser")
            result_block = soup.find_all("div", attrs={"class": "g"})
            if not result_block:
                print("No results found on this page")
                break
            
            print(f"Found {len(result_block)} results on this page")
            for result in result_block:
                link = result.find("a", href=True)
                title = result.find("h3")
                if link and title:
                    link = link["href"]
                    title = title.get_text()
                    print(f"Processing link: {link}")
                    try:
                        webpage = session.get(link, headers=headers, timeout=timeout)
                        webpage.raise_for_status()
                        visible_text = extract_text_from_webpage(webpage.text)
                        if len(visible_text) > max_chars_per_page:
                            visible_text = visible_text[:max_chars_per_page] + "..."
                        all_results.append({"link": link, "title": title, "text": visible_text})
                        print(f"Successfully extracted text from {link}")
                    except requests.exceptions.RequestException as e:
                        print(f"Error retrieving webpage content: {e}")
                        all_results.append({"link": link, "title": title, "text": None})
                else:
                    print("No link or title found for this result")
                    all_results.append({"link": None, "title": None, "text": None})
            start += len(result_block)

    print(f"Search completed. Total results: {len(all_results)}")
    print("Search results:")
    for i, result in enumerate(all_results, 1):
        print(f"Result {i}:")
        print(f"  Title: {result['title']}")
        print(f"  Link: {result['link']}")
        if result['text']:
            print(f"  Text: {result['text'][:100]}...")  # Print first 100 characters
        else:
            print("  Text: None")
    print("End of search results")

    if not all_results:
        print("No search results found. Returning a default message.")
        return [{"link": None, "title": "No Results", "text": "No information found in the web search results."}]

    return all_results

def summarize_content(content, model):
    if content is None:
        return "No content available to summarize."

    # Approximate the token limit using character count
    # Assuming an average of 4 characters per token
    max_chars = 7000 * 4  # Leave some room for the prompt
    if len(content) > max_chars:
        content = content[:max_chars] + "..."
    
    summary_prompt = f"""
    Summarize the following content concisely:
    {content}
    Summary:
    """
    summary = generate_chunked_response(model, summary_prompt, max_tokens=200)
    return summary

def rank_search_results(titles, summaries, model):
    if not titles or not summaries:
        print("No titles or summaries to rank.")
        return list(range(1, len(titles) + 1))

    ranking_prompt = (
        "Rank the following search results from a financial analyst perspective. "
        f"Assign a rank from 1 to {len(titles)} based on relevance, with 1 being the most relevant. "
        "Return only the numeric ranks in order, separated by commas.\n\n"
        "Titles and summaries:\n"
    )
    
    for i, (title, summary) in enumerate(zip(titles, summaries), 1):
        ranking_prompt += f"{i}. Title: {title}\nSummary: {summary}\n\n"
    
    ranking_prompt += "Ranks:"
    
    try:
        ranks_str = generate_chunked_response(model, ranking_prompt)
        ranks = [float(rank.strip()) for rank in ranks_str.split(',') if rank.strip()]
        
        # Check if we have the correct number of ranks
        if len(ranks) != len(titles):
            print(f"Warning: Number of ranks ({len(ranks)}) does not match number of titles ({len(titles)})")
            print(f"Model output: {ranks_str}")
            return list(range(1, len(titles) + 1))
        
        return ranks
    except Exception as e:
        print(f"Error in ranking: {str(e)}. Using fallback ranking method.")
        return list(range(1, len(titles) + 1))

def ask_question(question, temperature, top_p, repetition_penalty, web_search):
    global conversation_history

    if not question:
        return "Please enter a question."

    model = get_model(temperature, top_p, repetition_penalty)
    embed = get_embeddings()

    # Check if the FAISS database exists
    if os.path.exists("faiss_database"):
        database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
    else:
        database = None

    if web_search:
        search_results = google_search(question)
        
        processed_results = []
        for index, result in enumerate(search_results, start=1):
            if result["text"] is not None:
                try:
                    summary = summarize_content(result["text"], model)
                    processed_results.append({
                        "title": result.get("title", f"Result {index}"),
                        "content": result["text"],
                        "summary": summary,
                        "index": index
                    })
                except Exception as e:
                    print(f"Error processing search result {index}: {str(e)}")
            else:
                print(f"Skipping result {index} due to None content")
        
        if not processed_results:
            return "No valid search results found."

        # Rank the results
        titles = [r["title"] for r in processed_results]
        summaries = [r["summary"] for r in processed_results]
        try:
            ranks = rank_search_results(titles, summaries, model)
        except Exception as e:
            print(f"Error in ranking results: {str(e)}. Using default ranking.")
            ranks = list(range(1, len(processed_results) + 1))

        # Update Vector DB
        current_date = datetime.now().strftime("%Y-%m-%d")
        update_vector_db_with_search_results(processed_results, ranks, current_date)
        
        # Prepare context for the question
        context_str = "\n\n".join([f"Title: {r['title']}\nSummary: {r['summary']}\nRank: {ranks[i]}" 
                                   for i, r in enumerate(processed_results)])
        
        prompt_template = """
        Answer the question based on the following web search results:
        Web Search Results:
        {context}
        Current Question: {question}
        If the web search results don't contain relevant information, state that the information is not available in the search results.
        Provide a concise and direct answer to the question without mentioning the web search or these instructions:
        """
        prompt_val = ChatPromptTemplate.from_template(prompt_template)
        formatted_prompt = prompt_val.format(context=context_str, question=question)
    else:
        if database is None:
            return "No documents available. Please upload documents or enable web search to answer questions."

        history_str = "\n".join([f"Q: {item['question']}\nA: {item['answer']}" for item in conversation_history])

        if is_related_to_history(question, conversation_history):
            context_str = "No additional context needed. Please refer to the conversation history."
        else:
            retriever = database.as_retriever()
            relevant_docs = retriever.get_relevant_documents(question)
            context_str = "\n".join([doc.page_content for doc in relevant_docs])

        prompt_val = ChatPromptTemplate.from_template(prompt)
        formatted_prompt = prompt_val.format(history=history_str, context=context_str, question=question)

    full_response = generate_chunked_response(model, formatted_prompt)
    
    # Extract only the part after the last occurrence of a prompt-like sentence
    answer_patterns = [
        r"Provide a concise and direct answer to the question without mentioning the web search or these instructions:",
        r"Provide a concise and direct answer to the question:",
        r"Answer:"
    ]
    
    for pattern in answer_patterns:
        match = re.split(pattern, full_response, flags=re.IGNORECASE)
        if len(match) > 1:
            answer = match[-1].strip()
            break
    else:
        # If no pattern is found, return the full response
        answer = full_response.strip()

    if not web_search:
        memory_database[question] = answer
        conversation_history = manage_conversation_history(question, answer, conversation_history)

    return answer

def update_vectors(files, use_recursive_splitter):
    if not files:
        return "Please upload at least one PDF file."
    
    embed = get_embeddings()
    total_chunks = 0
    
    all_data = []
    for file in files:
        if use_recursive_splitter:
            data = load_and_split_document_recursive(file)
        else:
            data = load_and_split_document_basic(file)
        all_data.extend(data)
        total_chunks += len(data)
    
    if os.path.exists("faiss_database"):
        database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
        database.add_documents(all_data)
    else:
        database = FAISS.from_documents(all_data, embed)
    
    database.save_local("faiss_database")
    
    return f"Vector store updated successfully. Processed {total_chunks} chunks from {len(files)} files."

def update_vector_db_with_search_results(search_results, summaries, ranks):
    embed = get_embeddings()
    database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True) if os.path.exists("faiss_database") else FAISS.from_documents([], embed)
    
    current_date = datetime.now().strftime("%Y-%m-%d")
    
    documents = []
    for result, summary, rank in zip(search_results, summaries, ranks):
        if summary:  # Only create a document if there's a summary
            doc = Document(
                page_content=summary,
                metadata={
                    "search_date": current_date,
                    "search_title": result["title"],
                    "search_content": result["text"],
                    "search_summary": summary,
                    "rank": rank
                }
            )
            documents.append(doc)
    
    if documents:  # Only update the database if there are documents to add
        database.add_documents(documents)
        database.save_local("faiss_database")
    else:
        print("No valid documents to add to the database.")

def export_vector_db_to_excel():
    embed = get_embeddings()
    database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
    
    documents = database.docstore._dict.values()
    data = [{
        "Search Date": doc.metadata["search_date"],
        "Search Title": doc.metadata["search_title"],
        "Search Content": doc.metadata["search_content"],
        "Search Summary": doc.metadata["search_summary"],
        "Rank": doc.metadata["rank"]
    } for doc in documents]
    
    df = pd.DataFrame(data)
    
    with NamedTemporaryFile(delete=False, suffix='.xlsx') as tmp:
        excel_path = tmp.name
        df.to_excel(excel_path, index=False)
    
    return excel_path
    
def extract_db_to_excel():
    embed = get_embeddings()
    database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
    
    documents = database.docstore._dict.values()
    data = [{"page_content": doc.page_content, "metadata": json.dumps(doc.metadata)} for doc in documents]
    df = pd.DataFrame(data)
    
    with NamedTemporaryFile(delete=False, suffix='.xlsx') as tmp:
        excel_path = tmp.name
        df.to_excel(excel_path, index=False)
    
    return excel_path

def export_memory_db_to_excel():
    data = [{"question": question, "answer": answer} for question, answer in memory_database.items()]
    df_memory = pd.DataFrame(data)
    
    data_history = [{"question": item["question"], "answer": item["answer"]} for item in conversation_history]
    df_history = pd.DataFrame(data_history)
    
    with NamedTemporaryFile(delete=False, suffix='.xlsx') as tmp:
        excel_path = tmp.name
        with pd.ExcelWriter(excel_path, engine='openpyxl') as writer:
            df_memory.to_excel(writer, sheet_name='Memory Database', index=False)
            df_history.to_excel(writer, sheet_name='Conversation History', index=False)
    
    return excel_path

# Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("# Chat with your PDF documents")
    
    with gr.Row():
        file_input = gr.Files(label="Upload your PDF documents", file_types=[".pdf"])
        update_button = gr.Button("Update Vector Store")
        use_recursive_splitter = gr.Checkbox(label="Use Recursive Text Splitter", value=False)
    
    update_output = gr.Textbox(label="Update Status")
    update_button.click(update_vectors, inputs=[file_input, use_recursive_splitter], outputs=update_output)
    
    with gr.Row():
        with gr.Column(scale=2):
            chatbot = gr.Chatbot(label="Conversation")
            question_input = gr.Textbox(label="Ask a question about your documents")
            submit_button = gr.Button("Submit")
        with gr.Column(scale=1):
            temperature_slider = gr.Slider(label="Temperature", minimum=0.0, maximum=1.0, value=0.5, step=0.1)
            top_p_slider = gr.Slider(label="Top P", minimum=0.0, maximum=1.0, value=0.9, step=0.1)
            repetition_penalty_slider = gr.Slider(label="Repetition Penalty", minimum=1.0, maximum=2.0, value=1.0, step=0.1)
            web_search_checkbox = gr.Checkbox(label="Enable Web Search", value=False)
    
    def chat(question, history, temperature, top_p, repetition_penalty, web_search):
        answer = ask_question(question, temperature, top_p, repetition_penalty, web_search)
        history.append((question, answer))
        return "", history
    
    submit_button.click(chat, inputs=[question_input, chatbot, temperature_slider, top_p_slider, repetition_penalty_slider, web_search_checkbox], outputs=[question_input, chatbot])
    
    export_vector_db_button = gr.Button("Export Vector DB to Excel")
    vector_db_excel_output = gr.File(label="Download Vector DB Excel File")
    export_vector_db_button.click(export_vector_db_to_excel, inputs=[], outputs=vector_db_excel_output)

    extract_button = gr.Button("Extract Database to Excel")
    excel_output = gr.File(label="Download Excel File")
    extract_button.click(extract_db_to_excel, inputs=[], outputs=excel_output)
    
    export_memory_button = gr.Button("Export Memory Database to Excel")
    memory_excel_output = gr.File(label="Download Memory Excel File")
    export_memory_button.click(export_memory_db_to_excel, inputs=[], outputs=memory_excel_output)
    
    clear_button = gr.Button("Clear Cache")
    clear_output = gr.Textbox(label="Cache Status")
    clear_button.click(clear_cache, inputs=[], outputs=clear_output)

if __name__ == "__main__":
    demo.launch()