app.py

import os
import time
from openai import OpenAI
import numpy as np
import streamlit as st
import tensorflow as tf
import tensorflow_text
# import plotly.graph_objects as go
# from dotenv import load_dotenv
from langchain_openai import OpenAI as OpenAiLC
from langchain.memory import ConversationSummaryMemory, ChatMessageHistory
from llm import sys_instruction


##############
# PAGE STYLES

# Set page title and icon
st.set_page_config(page_title="EmoInsight",
                   page_icon=":robot_face:",
                   initial_sidebar_state="expanded",)

# Custom css styles
with open('style.css') as f:
    st.markdown(f'<style>{f.read()}</style>', unsafe_allow_html=True)


# Load variables from .env file
# load_dotenv()

# Load large model


@st.cache_resource  # Decorator to cache non-data objects
def Loading_sentiment_analysis_model():
    model = tf.saved_model.load('one_2')
    return model


senti_model = Loading_sentiment_analysis_model()


emoji_mapping = {
    "sadness": "😢",
    "neutral": "😐",
    "joy": "😄",
    "anger": "😡",
    "fear": "😨",
    "love": "❤️",
    "surprise": "😲",
}

emotion_categories = {
    0: 'anger',
    1: 'fear',
    2: 'joy',
    3: 'love',
    4: 'neutral',
    5: 'sadness',
    6: 'surprise'
}


##################
# STATE VARIABLES

# set api key
if 'key' not in st.session_state:
    st.session_state.key = os.environ["API_TOKEN"]

# openai.api_key = st.session_state.key

# gpt llm
if 'llm' not in st.session_state:
    st.session_state.llm = OpenAiLC(
        temperature=0.2, openai_api_key=st.session_state.key)

# model name
if "openai_model" not in st.session_state:
    st.session_state["openai_model"] = "gpt-3.5-turbo"

# openai client 
# model name
if "client" not in st.session_state:
    st.session_state["client"]  = OpenAI(
    api_key=st.session_state.key
)
    
# st chat history
if "message_history" not in st.session_state:
    st.session_state.message_history = []

# set instruction for gpt response
if 'sys_inst' not in st.session_state:
    st.session_state.sys_inst = sys_instruction()

# dict to store user question emotion
if 'emotion_counts' not in st.session_state:
    st.session_state.emotion_counts = {
        'anger': 0,
        'fear': 0,
        'joy': 0,
        'love': 0,
        'neutral': 0,
        'sadness': 0,
        'surprise': 0
    }


#######################
# LANG-CHAIN VARIABLES

# storing chat history
if 'old_summary' not in st.session_state:
    st.session_state.old_summary = 'User came to psychological assistant chatbot'

# langChian msg history
if 'lg_msg_history' not in st.session_state:
    st.session_state.lg_msg_history = ChatMessageHistory()

# summarize old conversation
if 'memory' not in st.session_state:
    st.session_state.memory = ConversationSummaryMemory.from_messages(
        llm=st.session_state.llm,
        buffer=st.session_state.old_summary,
        return_messages=True,
        chat_memory=st.session_state.lg_msg_history)


#############################################
#                MAIN APP                   #
#############################################
st.sidebar.markdown('')
st.sidebar.markdown('')
st.sidebar.markdown('')
st.sidebar.success("Select `Sentiment Plot` button to see the Emotino Graph")
st.sidebar.markdown('')
clear_chats = st.sidebar.button('Clear Chat')

if clear_chats:
    st.session_state.lg_msg_history.clear()
    st.session_state.old_summary = 'User came to psychological assistant chatbot'
    st.session_state.message_history = []
    alert = st.sidebar.warning('Chat cleared', icon='🚨')
    time.sleep(2)  # Wait for 3 seconds
    alert.empty()  # Clear the alert


st.markdown("<h1><center>EmoInsight</center></h1>",
            unsafe_allow_html=True)


# greetings
if len(st.session_state.message_history) == 0:
    # add to st history
    st.session_state.message_history.append(
        {"role": "assistant", "content": "How can I help you?"})
    # add to lg history
    # st.session_state.lg_msg_history.add_ai_message("How can I help you?")

# HISTORY
for message in st.session_state.message_history:
    if message['role'] == 'system':
        with st.chat_message("Emotion", avatar=emoji_mapping.get(message["content"])):
            a = "Sentiment: {}".format(message["content"])
            st.markdown(a)
    else:
        with st.chat_message(message["role"]):
            st.markdown(message["content"])

# CHAT BOT
if prompt := st.chat_input("What is up?"):
    # USER
    with st.chat_message("user"):
        st.markdown(prompt)
    # add to st history
    st.session_state.message_history.append(
        {"role": "user", "content": prompt})
    # add to lg history
    st.session_state.lg_msg_history.add_user_message(prompt)

    # SENTIMENT PREDICION
    emotion = senti_model([prompt])
    true_classes = np.argmax(emotion, axis=1)
    emotion_category = emotion_categories.get(int(true_classes))
    st.session_state.emotion_counts[emotion_category] += 1

    # EMOTION
    with st.chat_message("Emotion", avatar=emoji_mapping.get(emotion_category)):
        st.write("Sentiment: {}".format(emotion_category))
        st.session_state.message_history.append(
            {"role": "system", "content": emotion_category})

    # AI BOT
    with st.chat_message("assistant"):
        message_placeholder = st.empty()
        full_response = ""

        # get response
        for chunk in st.session_state.client.chat.completions.create(
                model=st.session_state["openai_model"],
                messages=[
                    {"role": "system", "content": st.session_state.sys_inst.format(
                        history=st.session_state.old_summary)},
                    {"role": "user", "content": prompt}
                ],  # pass old chat history
                stream=True):

            # render gpt response in realtime
            if chunk.choices[0].delta.content:
                # print(chunk.choices[0].delta.content)   
                full_response += chunk.choices[0].delta.content
                message_placeholder.markdown(full_response + "▌")
        message_placeholder.markdown(full_response)

    # add to st history
    st.session_state.message_history.append(
        {"role": "assistant", "content": full_response})
    # add to lg history
    st.session_state.lg_msg_history.add_ai_message(prompt)

# Clear old chat after 4 dialogs
# And update old summary with new summary
chat_len = len(st.session_state.lg_msg_history.messages)
if (chat_len >= 4) and (chat_len % 4 == 0):

    # get new summary of chat
    st.session_state.old_summary = st.session_state.memory.predict_new_summary(
        messages=st.session_state.lg_msg_history.messages,
        existing_summary=st.session_state.old_summary)

    # flush old lg-chat history
    st.session_state.lg_msg_history.clear()