pages/chat.py

from openai import OpenAI
import streamlit as st
from langchain_openai import ChatOpenAI
from langchain_openai.embeddings import OpenAIEmbeddings
from langchain_text_splitters import RecursiveCharacterTextSplitter
import markdown
from operator import itemgetter
from langchain.schema.runnable import RunnablePassthrough
from langchain_core.prompts import ChatPromptTemplate
from langchain.schema import Document
from dotenv import load_dotenv
from langchain_community.vectorstores import Qdrant
# from langchain_qdrant import Qdrant
import os
import pandas as pd
import numpy as np

st.set_page_config(
    page_title="Narrativ 🧠",
    layout="wide",
    initial_sidebar_state="expanded",
    page_icon="🧠",
)

# Custom CSS for enhanced styling
st.markdown("""
    <style>
    .circular-image {
        width: 200px;
        height: 200px;
        border-radius: 50%;
        object-fit: cover;
        display: block;
        margin: 20px auto;
        box-shadow: 0 4px 8px rgba(0,0,0,0.1);
    }
    
    /* Container for search section */
    .search-container {
        background: white;
        padding: 20px;
        border-radius: 10px;
        box-shadow: 0 2px 4px rgba(0,0,0,0.1);
        margin: 20px 0;
    }
    
    /* Combined search input styling */
    .combined-search {
        display: flex;
        gap: 10px;
        align-items: center;
        margin-bottom: 20px;
    }
    </style>
    """, unsafe_allow_html=True)

load_dotenv()
OPENAI_API_KEY = os.environ["OPENAI_API_KEY"]
base_llm = ChatOpenAI(model="gpt-4o")
embedding_model = OpenAIEmbeddings(model="text-embedding-3-small")

#========== APP

from PIL import Image, ImageEnhance

image = Image.open('./data/Sentiment_index_traffic.png')
#enhancer = ImageEnhance.Brightness(image)
#darker_image = enhancer.enhance(0.5)  # Adjust the brightness factor as needed
st.image(image, output_format="PNG", clamp=True)

st.title("Narrativ 📰")

#check1 = st.button("Submit", key="submit_button")
prompt='traffic'

prompt=st.session_state.prompt
date=st.session_state.date
# Change the sidebar background with enhanced gradient and text styling
# sideb.markdown(

if 'messages' not in st.session_state:
    st.session_state.messages = []

st.session_state.messages.append({"role": "assistant", "content": f'{date} {prompt}'})


if prompt:
    if date:
        try:
            data=pd.read_csv('./data/sentiment_index_traffic_index_final1.csv',
                index_col='index',
                parse_dates=True,
                infer_datetime_format=True
            ).drop(columns=['llm_index','sentiment_index_hf','confidence_hf'])

            data = data.loc[data.index == date]
            filtered_data = data[data.apply(lambda row: row.astype(str).str.contains(prompt, na=False).any(), axis=1)]
            urls = data['url'].values.flatten()
            data_all = filtered_data.values.flatten()
            docs = data_all 
            if len(docs)==0:
                st.warning("No articles found that contain the topic on the given day.")

        except Exception as e:
            st.error(f"Error processing date: {e}")
    else:
        try:
            data = pd.read_csv(
                './data/sentiment_index_traffic_index_final1.csv',
                index_col='index',
                parse_dates=True,
                infer_datetime_format=True
            ).drop(columns=['llm_index','sentiment_index_hf','confidence_hf'])
            
            filtered_data = data[data.apply(lambda row: row.astype(str).str.contains(prompt, na=False).any(), axis=1)]
            data_all = filtered_data.values.flatten()
            urls = data['url'].values.flatten()
            docs = data_all 
            if len(docs)==0:
                st.warning("No articles found that contain the topic on the given day.")
            
            #data_all = data.values.flatten()
            #docs = data_all
            # with open(f'./data/sentiment_index_traffic_index_final1.md', "w", encoding="utf-8") as file:
            #     file.write(str(data_all))
            # with open(f'./data/sentiment_index_traffic_index_final1.md', "r", encoding="utf-8") as file_content:
            #     docs = file_content.read()
        except Exception as e:
            st.error(f"Error loading data: {e}")

else:
    data=pd.read_csv('./data/sentiment_index_traffic_index_final1.csv',
                index_col='index',
                parse_dates=True,
                infer_datetime_format=True
            ).drop(columns=['llm_index','sentiment_index_hf','confidence_hf'])

    urls = data['url'].values.flatten()
    data_all = data.values.flatten()
    docs = data_all 
    if len(docs)==0:
        st.warning("No articles found that contain the topic on the given day.")
                
if len(docs)>0:
    docs_text = "\n".join([f"- {value}" for value in data_all if not pd.isna(value)])
    docs = [Document(page_content=docs_text)]
    st.write(data)

    try:
        no_rows=len(data)
        data_summary=data.groupby('title').first()
        filtered_data = data_summary[data_summary.apply(lambda row: row.astype(str).str.contains(prompt, na=False).any(), axis=1)]
        data_all_summary = filtered_data['summary_date'].groupby('title').first().values.flatten()
        docs_text_summary = "\n".join([f"- {value}" for value in data_all_summary if not pd.isna(value)])
        summary_data=str(docs_text_summary) #docs['summary_date']
        print('heere',summary_data)
        summary = base_llm.invoke(f"""format nicely the summary into paragraphs for Streamlit. Say how many news articles are available for the given data, the number is: {no_rows}.
                                ## Output Format:
    - **Summary of Opinions:** [Concise summary of key opinions]
    - **Sentiment Analysis:**
    - Sentiment: [Positive/Negative/Neutral]
    - Reasoning: [Detailed explanation here]
    - **Chain-of-Thought Reasoning:** [Step-by-step explanation]
    - **Sources:** [URLs for 5 most critical and recent articles on this topic]
    ## Guidelines:
    - Maintain objectivity and precision in your analysis.
    - Focus on the context specific to the Greater Washington Area.
    - Use professional and analytical language suitable for client reports.
    - Respond in the language of the article (mostly English).
    - From the provided context, add the URL sources, you find them here, URLs: {urls} - make sure they are clicable! related to the topic.
                                Context: {summary_data}""").content #lcel_rag_chain.invoke({"question": prompt})
        print(summary)
        st.chat_message("assistant").write((summary))
        st.session_state.messages.append({"role": "assistant", "content": summary})
    except Exception as e:
        st.error(f"Error generating summary: {e}")

    if date:
        with open('./data/sentiment_index_traffic_index_final_date.md', 'w') as file:
                file.write(str(data_all))
    else:
        with open('./data/sentiment_index_traffic_index_final1.md', 'w') as file:
                file.write(str(data_all))


    client = OpenAI(api_key=OPENAI_API_KEY)
    
    if "openai_model" not in st.session_state:
        st.session_state["openai_model"] = "gpt-4o"
    
    prompt1 = st.chat_input("Type your additional questions here...")
    
    # Suggested keywords with enhanced styling
    suggested_keywords = ["Summarize results", f"Explain the traffic drop", f"Explain the traffic growth"]
    st.markdown("**Suggested Keywords:**")
    cols = st.columns(len(suggested_keywords))
    for idx, keyword in enumerate(suggested_keywords):
        if cols[idx].button(keyword, key=keyword):
            prompt1 = keyword
    
    if prompt1:
        if date:
            file_path = f'./data/sentiment_index_traffic_index_final_date.md'
        else:
            file_path = f'./data/sentiment_index_traffic_index_final1.md'
            
        try:
            with open(file_path, "r", encoding="utf-8") as file_content:
                docs = file_content.read()
        except Exception as e:
            st.error(f"Error loading context: {e}")
            docs = ""

        # Add user message to chat history
        st.session_state.messages.append({"role": "user", "content": f'You are a Transurban traffic analyst, that focuses on the Express lanes I-495 and I-95 in the Greater Washington Area. Having this knowledge answer questions: {prompt1} using context from {docs}'})
        # Display user message in chat message container
        with st.chat_message("user"):
            st.markdown(prompt1)
        # Display assistant response in chat message container
        with st.chat_message("assistant"):
            try:
                stream = client.chat.completions.create(
                    model=st.session_state["openai_model"],
                    messages=[
                        {"role": m["role"], "content": m["content"]}
                        for m in st.session_state.messages
                    ],
                    stream=True,
                )
                response = st.write_stream(stream)
                st.session_state.messages.append({"role": "assistant", "content": response})
            except Exception as e:
                st.error(f"Error generating response: {e}")