app.py

import streamlit as st
import random
import pandas as pd
import requests
from io import BytesIO
from PIL import Image
from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM
import re
import time

# --------------------------- Configuration & Session State ---------------------------
# Define maximum dimensions for the fortune image (in pixels)
MAX_SIZE = (400, 400)

# Initialize button click count in session state
if "button_count_temp" not in st.session_state:
    st.session_state.button_count_temp = 0

# Set page configuration and title
st.set_page_config(page_title="Fortune Stick Enquiry", layout="wide")
st.title("Fortune Stick Enquiry")

# Initialize session state variables for managing application state
if "submitted_text" not in st.session_state:
    st.session_state.submitted_text = False
if "fortune_number" not in st.session_state:
    st.session_state.fortune_number = None
if "fortune_row" not in st.session_state:
    st.session_state.fortune_row = None
if "error_message" not in st.session_state:
    st.session_state.error_message = ""
if "cfu_explain_text" not in st.session_state:
    st.session_state.cfu_explain_text = ""
if "stick_clicked" not in st.session_state:
    st.session_state.stick_clicked = False

# Load fortune details from CSV file into session state
if "fortune_data" not in st.session_state:
    try:
        st.session_state.fortune_data = pd.read_csv("/home/user/app/resources/detail.csv")
    except Exception as e:
        st.error(f"Error loading CSV: {e}")
        st.session_state.fortune_data = None

# --------------------------- Model Functions ---------------------------
# Function to load a fine-tuned classifier model and predict a label based on the question
def load_finetuned_classifier_model(question):
    label_list = ["Geomancy", "Lost Property", "Personal Well-Being", "Future Prospect", "Traveling"]
    # Mapping to convert default "LABEL_x" outputs to meaningful labels
    mapping = {f"LABEL_{i}": label for i, label in enumerate(label_list)}

    pipe = pipeline("text-classification", model="tonyhui2234/CustomModel_classifier_model_10")
    prediction = pipe(question)[0]['label']
    predicted_label = mapping.get(prediction, prediction)
    return predicted_label

# Function to generate a detailed answer by combining the user's question and the fortune detail
def generate_answer(question, fortune):
    # Start measuring runtime
    start_time = time.perf_counter()
    tokenizer = AutoTokenizer.from_pretrained("tonyhui2234/finetuned_model_text_gen")
    model = AutoModelForSeq2SeqLM.from_pretrained("tonyhui2234/finetuned_model_text_gen", device_map="auto")
    input_text = "Question: " + question + " Fortune: " + fortune
    inputs = tokenizer(input_text, return_tensors="pt", truncation=True)
    outputs = model.generate(
        **inputs,
        max_length=256,
        num_beams=4,
        early_stopping=True,
        repetition_penalty=2.0,
        no_repeat_ngram_size=3
    )
    answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
    
    # Stop measuring runtime
    run_time = time.perf_counter() - start_time
    print(f"Runtime: {run_time:.4f} seconds")
    return answer

# Function that combines analysis with regex to extract the related fortune detail and then generate an answer
def analysis(row_detail, classifiy, question):
    # Use the classifier's output to match the corresponding detail in the fortune data
    pattern = re.compile(re.escape(classifiy) + r":\s*(.*?)(?:\.|$)", re.IGNORECASE)
    match = pattern.search(row_detail)
    if match:
        result = match.group(1)
        # Generate a custom answer based on the matched fortune detail and the user's question
        return generate_answer(question, result)
    else:
        return "Heaven's secret cannot be revealed."

# Function to check if the input sentence is in English using a language detection model
def check_sentence_is_english_model(question):
    pipe_english = pipeline("text-classification", model="eleldar/language-detection")
    return pipe_english(question)[0]['label'] == 'en'

# Function to check if the input sentence is a question using a question vs. statement classifier
def check_sentence_is_question_model(question):
    pipe_question = pipeline("text-classification", model="shahrukhx01/question-vs-statement-classifier")
    return pipe_question(question)[0]['label'] == 'LABEL_1'

# --------------------------- Callback Functions ---------------------------
# Callback for when the submit button is clicked
def submit_text_callback():    
    question = st.session_state.get("user_sentence", "")
    # Clear any previous error message
    st.session_state.error_message = ""
    
    # Validate that the input is in English and is a question
    if not check_sentence_is_english_model(question):
        st.session_state.error_message = "Please enter in English!"
        st.session_state.button_count_temp = 0
        return

    if not check_sentence_is_question_model(question):
        st.session_state.error_message = "This is not a question. Please enter again!"
        st.session_state.button_count_temp = 0
        return

    # Require a second confirmation click to proceed
    if st.session_state.button_count_temp == 0:
        st.session_state.error_message = "Please take a moment to quietly reflect on your question in your mind, then click submit again!"
        st.session_state.button_count_temp = 1
        return

    # If validations pass, set submission flag and reset click counter
    st.session_state.submitted_text = True
    st.session_state.button_count_temp = 0

    # Randomly generate a fortune number between 1 and 100
    st.session_state.fortune_number = random.randint(1, 100)
    
    # Retrieve corresponding fortune details from the CSV based on the generated number
    df = st.session_state.fortune_data
    row_detail = ''
    if df is not None:
        matching_row = df[df['CNumber'] == st.session_state.fortune_number]
        if not matching_row.empty:
            row = matching_row.iloc[0]
            row_detail = row.get("Detail", "No detail available.")
            st.session_state.fortune_row = {
                "Header": row.get("Header", "N/A"),
                "Luck": row.get("Luck", "N/A"),
                "Description": row.get("Description", "No description available."),
                "Detail": row_detail,
                "HeaderLink": row.get("link", None)
            }
        else:
            st.session_state.fortune_row = {
                "Header": "N/A",
                "Luck": "N/A",
                "Description": "No description available.",
                "Detail": "No detail available.",
                "HeaderLink": None
            }

# Function to load and resize a local image file
def load_and_resize_image(path, max_size=MAX_SIZE):
    try:
        img = Image.open(path)
        img.thumbnail(max_size, Image.Resampling.LANCZOS)
        return img
    except Exception as e:
        st.error(f"Error loading image: {e}")
        return None

# Function to download an image from a URL and resize it
def download_and_resize_image(url, max_size=MAX_SIZE):
    try:
        response = requests.get(url)
        response.raise_for_status()
        image_bytes = BytesIO(response.content)
        img = Image.open(image_bytes)
        img.thumbnail(max_size, Image.Resampling.LANCZOS)
        return img
    except Exception as e:
        st.error(f"Error loading image from URL: {e}")
        return None

# Callback for when the 'Cfu Explain' button is clicked
def stick_enquiry_callback():
    # Retrieve the user's question and ensure fortune data is available
    question = st.session_state.get("user_sentence", "")
    if not st.session_state.fortune_row:
        st.error("Fortune data is not available. Please submit your question first.")
        return
    row_detail = st.session_state.fortune_row.get("Detail", "No detail available.")
    
    # Classify the question to determine which fortune detail to use
    classifiy = load_finetuned_classifier_model(question)
    # Generate an explanation based on the classification and fortune detail
    cfu_explain = analysis(row_detail, classifiy, question)
    # Save the generated explanation for display
    st.session_state.cfu_explain_text = cfu_explain
    st.session_state.stick_clicked = True

# --------------------------- Layout & Display ---------------------------
# Define the main layout with two columns: left for user input and right for fortune display
left_col, _, right_col = st.columns([3, 1, 5])

# ---- Left Column: User Input and Interaction ----
with left_col:
    left_top = st.container()
    left_bottom = st.container()
    
    # Top container: Question input and submission button
    with left_top:
        st.text_area("Enter your question in English", key="user_sentence", height=150)
        st.button("submit", key="submit_button", on_click=submit_text_callback)
        if st.session_state.error_message:
            st.error(st.session_state.error_message)
    
    # Bottom container: Button to trigger explanation and display the generated answer
    if st.session_state.submitted_text:
        with left_bottom:
            # Add spacing for better visual separation
            for _ in range(5):
                st.write("")
            col1, col2, col3 = st.columns(3)
            with col2:
                st.button("Cfu Explain", key="stick_button", on_click=stick_enquiry_callback)
            if st.session_state.stick_clicked:
                # Display the generated explanation text
                st.text_area(' ', value=st.session_state.cfu_explain_text, height=300, disabled=True)

# ---- Right Column: Fortune Display (Image and Details) ----
with right_col:
    with st.container():
        col_left, col_center, col_right = st.columns([1, 2, 1])
        with col_center:
            # Display fortune image based on fortune data availability
            if st.session_state.submitted_text and st.session_state.fortune_row:
                header_link = st.session_state.fortune_row.get("HeaderLink")
                if header_link:
                    img_from_url = download_and_resize_image(header_link)
                    if img_from_url:
                        st.image(img_from_url, use_container_width=False)
                    else:
                        img = load_and_resize_image("/home/user/app/resources/error.png")
                        if img:
                            st.image(img, use_container_width=False)
                else:
                    img = load_and_resize_image("/home/user/app/resources/error.png")
                    if img:
                        st.image(img, use_container_width=False)
            else:
                img = load_and_resize_image("/home/user/app/resources/fortune.png")
                if img:
                    st.image(img, caption="Your Fortune", use_container_width=False)
    with st.container():
        # Display fortune details: Number, Luck, Description, and Detail
        if st.session_state.fortune_row:
            luck_text = st.session_state.fortune_row.get("Luck", "N/A")
            description_text = st.session_state.fortune_row.get("Description", "No description available.")
            detail_text = st.session_state.fortune_row.get("Detail", "No detail available.")
            
            summary = f"""
            <div style="font-size: 28px; font-weight: bold;">
                Fortune stick number: {st.session_state.fortune_number}<br>
                Luck: {luck_text}
            </div>
            """
            st.markdown(summary, unsafe_allow_html=True)
            
            st.text_area("Description", value=description_text, height=150, disabled=True)
            st.text_area("Detail", value=detail_text, height=150, disabled=True)