functions.py

import streamlit as st
import os, time
from llama_index.readers.schema.base import Document
from llama_index import LLMPredictor, GPTSimpleVectorIndex, PromptHelper, GPTTreeIndex
from langchain import OpenAI
import pandas as pd


def generate_response(prompt, index, llm_predictor, test=False):
    if test:
        return prompt
    else:
        response = index.query(prompt, llm_predictor=llm_predictor)
        return response


def update_chat_state():
    st.session_state.chat_sent = st.session_state.chat_input
    st.session_state.chat_input = ''


def get_chat_input():
    st.text_input(label="Write your query here", key="chat_input", on_change=update_chat_state)
    return st.session_state.chat_sent

def config_llm_predictor():
    llm_predictor = LLMPredictor(llm=OpenAI(temperature=0.3, model_name="gpt-3.5-turbo"))

    max_input_size = 8192
    num_output = None
    max_chunk_overlap = 20

    prompt_helper = PromptHelper(max_input_size = max_input_size, num_output = num_output, max_chunk_overlap = max_chunk_overlap)

    return llm_predictor


@st.cache(allow_output_mutation=True)
def load_index(json_index):
    #index = GPTTreeIndex.load_from_disk(json_index)
    index = GPTSimpleVectorIndex.load_from_disk(json_index)

    return index


@st.cache(allow_output_mutation=True)
def get_data():
 data = pd.read_csv("data/appstore_reviews.csv")
 data = data[["application", "review", "rating", "date"]]
 data["application"] = data["application"].str.lower()
 data = data[~data["review"].isna()]
 return data


def set_api_key(api_key_file):
    with open(api_key_file, "r") as file:
        openai_key = file.read().replace("\n", "")
    # Set environment
    os.environ["OPENAI_API_KEY"] = openai_key


def get_search(data):
    input_text = st.text_input("Search in reviews:", key="search")
    output = data[data["review"].apply(lambda x: x.lower()).str.contains(input_text.lower())]

    return output