app.py

import streamlit as st
import pandas as pd
import openai
import joblib
from PIL import Image
import requests
from io import BytesIO
import matplotlib.pyplot as plt
import numpy as np
from sklearn.preprocessing import LabelEncoder
from huggingface_hub import hf_hub_download
from transformers import pipeline
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
import re

# Dataset loading function with caching
@st.cache_data
def load_datasets():
    try:
        with st.spinner('Loading dataset...'):
            original_data = pd.read_csv('CTP_Model1.csv', low_memory=False)
            original_data.columns = original_data.columns.str.strip().str.capitalize()
            return original_data
    except Exception as e:
        st.error(f"Error loading dataset: {str(e)}")
        raise e

def load_image(image_file):
    return Image.open(image_file)

def classify_image(image):
    try:
        # Create a pipeline for image classification
        classifier = pipeline('image-classification', model="dima806/car_models_image_detection", device=-1)  # Use -1 for CPU, or 0 for GPU if available
        
        # Classify the image
        results = classifier(image)
        
        # Return top 5 predictions
        return results[:5]
    
    except Exception as e:
        st.error(f"Classification error: {e}")
        return None

def find_closest_match(df, brand, model):
    # Combine brand and model names from the dataset
    df['full_name'] = df['Make'] + ' ' + df['Model']
    
    # Create a list of all car names
    car_names = df['full_name'].tolist()
    
    # Add the query car name
    query_car = f"{brand} {model}"
    car_names.append(query_car)
    
    # Create TF-IDF vectorizer
    vectorizer = TfidfVectorizer()
    tfidf_matrix = vectorizer.fit_transform(car_names)
    
    # Compute cosine similarity
    cosine_similarities = cosine_similarity(tfidf_matrix[-1], tfidf_matrix[:-1]).flatten()
    
    # Get the index of the most similar car
    most_similar_index = cosine_similarities.argmax()
    
    # Return the most similar car's data
    return df.iloc[most_similar_index]

def get_car_overview(car_data):
    prompt = f"Provide an overview of the following car:\nYear: {car_data['Year']}\nMake: {car_data['Make']}\nModel: {car_data['Model']}\nTrim: {car_data['Trim']}\nPrice: ${car_data['Price']}\nCondition: {car_data['Condition']}\n"
    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages=[{"role": "user", "content": prompt}]
    )
    return response.choices[0].message['content']

def load_model_and_encodings():
    try:
        with st.spinner('Loading model...'):
            model_content = hf_hub_download(repo_id="EdBoy2202/car_prediction_model", filename="car_price_modelv3.pkl")
            model = joblib.load(model_content)

        original_data = load_datasets()
        
        label_encoders = {}
        categorical_features = ['Make', 'Model', 'Condition', 'Fuel', 'Title_status', 
                                'Transmission', 'Drive', 'Size', 'Type', 'Paint_color']
        
        for feature in categorical_features:
            if feature in original_data.columns:
                le = LabelEncoder()
                unique_values = original_data[feature].fillna('unknown').str.strip().unique()
                le.fit(unique_values)
                label_encoders[feature.lower()] = le
        
        return model, label_encoders
    except Exception as e:
        st.error(f"Error loading model: {str(e)}")
        raise e

def predict_price(model, encoders, user_input):
    encoded_features = {feature: encoders[feature].transform([value])[0] if value in encoders[feature] else 0 
                        for feature, value in user_input.items()}
    input_data = pd.DataFrame([encoded_features])
    predicted_price = model.predict(input_data)
    return predicted_price[0]

# Streamlit App
st.title("Auto Appraise")
st.write("Capture a car image using your camera or upload an image to get its brand, model, overview, and expected price!")

# Load model and encoders
model, label_encoders = load_model_and_encodings()

# Initialize OpenAI API key
openai.api_key = st.secrets["GPT_TOKEN"]

# Camera input for taking photo
camera_image = st.camera_input("Take a picture of the car!")

if camera_image is not None:
    image = load_image(camera_image)
    st.image(image, caption='Captured Image.', use_container_width=True)

    # Classify the car image
    with st.spinner('Analyzing image...'):
        car_classifications = classify_image(image)
    
    if car_classifications:
        st.subheader("Car Classification Results:")
        for classification in car_classifications:
            st.write(f"Model: {classification['label']}")
            st.write(f"Confidence: {classification['score']*100:.2f}%")
        
        # Use the top prediction for further processing
        top_prediction = car_classifications[0]['label']
        brand, model_name = top_prediction.split(' ', 1)
        
        st.write(f"Identified Car: {brand} {model_name}")

        # Find the closest match in the CSV
        df = load_datasets()
        match = find_closest_match(df, brand, model_name)
        if match is not None:
            st.write("Closest Match Found:")
            st.write(f"Make: {match['Make']}")
            st.write(f"Model: {match['Model']}")
            st.write(f"Year: {match['Year']}")
            st.write(f"Price: ${match['Price']}")

            # Get additional information using GPT-3.5-turbo
            overview = get_car_overview(match)
            st.write("Car Overview:")
            st.write(overview)

            # Interactive Price Prediction
            st.subheader("Price Prediction Over Time")
            selected_years = st.slider("Select range of years for price prediction", 
                                         min_value=2000, max_value=2023, value=(2010, 2023))

            years = np.arange(selected_years[0], selected_years[1] + 1)
            predicted_prices = []

            for year in years:
                user_input = {
                    'Make': match['Make'],
                    'Model': match['Model'],
                    'Condition': match['Condition'],
                    'Fuel': match['Fuel'],
                    'Title_status': match['Title_status'],
                    'Transmission': match['Transmission'],
                    'Drive': match['Drive'],
                    'Size': match['Size'],
                    'Type': match['Type'],
                    'Paint_color': match['Paint_color'],
                    'Year': year
                }
                
                price = predict_price(model, label_encoders, user_input)
                predicted_prices.append(price)

            # Plotting the results
            plt.figure(figsize=(10, 5))
            plt.plot(years, predicted_prices, marker='o')
            plt.title(f"Predicted Price of {match['Make']} {match['Model']} Over Time")
            plt.xlabel("Year")
            plt.ylabel("Predicted Price ($)")
            plt.grid()
            st.pyplot(plt)

        else:
            st.write("No match found in the database.")
    else:
        st.error("Could not classify the image. Please try again with a different image.")
else:
    st.write("Please take a picture of the car to proceed.")