import gradio as gr
import numpy as np
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.chains import LLMChain
from langchain import PromptTemplate
import re
import pandas as pd
from langchain.vectorstores import FAISS
import requests
from typing import List
from langchain.schema import (
    SystemMessage,
    HumanMessage,
    AIMessage
)
import os
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.chat_models import ChatOpenAI

from langchain.llms.base import LLM
from typing import Optional, List, Mapping, Any

import ast
from utils import ClaudeLLM, extract_website_name, remove_numbers

embeddings = HuggingFaceEmbeddings()
db = FAISS.load_local('db_full', embeddings)

mp_docs = {}
# llm = ClaudeLLM()
# ChatOpenAI(
#             temperature=0,
#             model='gpt-3.5-turbo-16k'
#         )


def add_text(history, text):

    print(history)
    history = history + [(text, None)]

    return history, ""

# pipeline = {'claude': (ClaudeLLM(), 0), 'gpt-3.5': (ChatOpenAI(temperature=0,model='gpt-3.5-turbo-16k'), 65), 'gpt-4': (ChatOpenAI(temperature=0, model='gpt-4'), 30)}

def retrieve_thoughts(query, n):

    # print(db.similarity_search_with_score(query = query, k = k, fetch_k = k*10))
    docs_with_score = db.similarity_search_with_score(query = query, k = len(db.index_to_docstore_id.values()), fetch_k = len(db.index_to_docstore_id.values()))
    df = pd.DataFrame([dict(doc[0])['metadata'] for doc in docs_with_score], )
    df = pd.concat((df, pd.DataFrame([dict(doc[0])['page_content'] for doc in docs_with_score], columns = ['page_content'])), axis = 1)
    df = pd.concat((df, pd.DataFrame([doc[1] for doc in docs_with_score], columns = ['score'])), axis = 1)

  # TO-DO: What if user query doesn't match what we provide as documents

    tier_1 = df[df['score'] < 0.7]
    tier_2 = df[(df['score'] < 0.95) * (df["score"] > 0.7)]


    chunks_1 = tier_1.groupby(['title', 'url', '_id']).apply(lambda x: "\n...\n".join(x.sort_values('id')['page_content'].values)).values
    tier_1_adjusted = tier_1.groupby(['title', 'url', '_id']).first().reset_index()[['_id', 'title', 'url', 'score']]
    tier_1_adjusted['ref'] = range(1, len(tier_1_adjusted) + 1 )
    tier_1_adjusted['content'] = chunks_1

    chunks_2 = tier_2.groupby(['title', 'url', '_id']).apply(lambda x: "\n...\n".join(x.sort_values('id')['page_content'].values)).values
    tier_2_adjusted = tier_2.groupby(['title', 'url', '_id']).first().reset_index()[['_id', 'title', 'url']]
    tier_2_adjusted['content'] = chunks_2

    if n:
      tier_1_adjusted = tier_1_adjusted[:min(len(tier_1_adjusted), n)]

    print(len(tier_1_adjusted))
  # tier_1 = [doc[0]  for doc in docs if ((doc[1] < 1))][:5]
  # tier_2 = [doc[0]  for doc in docs if ((doc[1] > 0.7)*(doc[1] < 1.5))][10:15]

    return {'tier 1':tier_1_adjusted, 'tier 2': tier_2.loc[:5]}

def qa_retrieve(query, llm):

    # llm = pipeline["claude"][0]

    docs = ""

    global db
    print(db)

    global mp_docs
    thoughts = retrieve_thoughts(query, 0)
    if not(thoughts):

        if mp_docs:
            thoughts = mp_docs
    else:
        mp_docs = thoughts

    tier_1 = thoughts['tier 1']
    tier_2 = thoughts['tier 2']
    
    reference = tier_1[['ref', 'url', 'title', 'content','score']].to_dict('records')

    # tier_1 = list(tier_1.apply(lambda x: f"[{int(x['ref'])}] title: {x['title']}\n Content: {x.content}", axis = 1).values)
    # print(len(tier_1))
    # tier_2 = list(tier_2.apply(lambda x: f"title: {x['title']}\n Content: {x.content}", axis = 1).values)

    return {'Reference': reference}

def flush():
  return None

examples = [
    ["Will Russia win the war in Ukraine?"],

    ]

demo = gr.Interface(fn=qa_retrieve, title="cicero-qa-api",
                     inputs=gr.inputs.Textbox(lines=5, label="what would you like to learn about?"),
                     outputs=[
                             gr.components.JSON( label="Reference")],examples=examples)

demo.queue(concurrency_count = 4)
demo.launch()