src/streamlit_app.py

import streamlit as st
import pandas as pd
import numpy as np
import joblib

# --- PAGE CONFIGURATION ---
st.set_page_config(
    page_title="Steel Plate Defect Prediction",
    page_icon="🏗️",
    layout="wide")

@st.cache_resource
def load_artifacts():
    artifacts = joblib.load('src/steel_defect_model.pkl')
    return artifacts

try:
    artifacts = load_artifacts()
    models = artifacts['models']
    scaler = artifacts['scaler']
    feature_names = artifacts['features']
    targets = artifacts['targets']
except FileNotFoundError:
    st.error("Model file 'steel_defect_model.pkl' not found.")
    st.stop()

st.title("🏗️ Steel Plate Defect Prediction AI")
st.markdown("""
This AI model predicts the probability of **7 different types of defects** in steel plates based on their geometric and radiometric properties.
""")

st.sidebar.header("Input Parameters")
st.sidebar.info("Adjust the sliders to simulate different steel plate properties.")

def user_input_features():
    data = {}
    
    st.sidebar.subheader("1. Geometry")
    data['X_Minimum'] = st.sidebar.number_input('X Minimum', 0, 1700, 0)
    data['X_Maximum'] = st.sidebar.number_input('X Maximum', 0, 1700, 50)
    data['Y_Minimum'] = st.sidebar.number_input('Y Minimum', 0, 13000000, 600000)
    data['Y_Maximum'] = st.sidebar.number_input('Y Maximum', 0, 13000000, 600050)
    data['Pixels_Areas'] = st.sidebar.number_input('Pixels Areas', 0, 20000, 200)
    data['Steel_Plate_Thickness'] = st.sidebar.slider('Steel Plate Thickness', 40, 300, 80)
    
    st.sidebar.subheader("2. Luminosity")
    data['Sum_of_Luminosity'] = st.sidebar.number_input('Sum of Luminosity', 0, 12000000, 20000)
    data['Minimum_of_Luminosity'] = st.sidebar.slider('Minimum Luminosity', 0, 200, 80)
    data['Maximum_of_Luminosity'] = st.sidebar.slider('Maximum Luminosity', 0, 260, 130)
    
    data['Length_of_Conveyer'] = 1459  
    data['TypeOfSteel_A300'] = st.sidebar.selectbox('Type of Steel A300', [0, 1], index=0)
    data['Edges_Index'] = 0.35
    data['Empty_Index'] = 0.4
    data['Square_Index'] = 0.57
    data['Outside_X_Index'] = 0.03
    data['Edges_X_Index'] = 0.61
    data['Edges_Y_Index'] = 0.83
    data['Outside_Global_Index'] = 0.5
    data['LogOfAreas'] = np.log(data['Pixels_Areas']) if data['Pixels_Areas'] > 0 else 0 
    data['Log_X_Index'] = 0 
    data['Log_Y_Index'] = 0 
    data['Orientation_Index'] = 0.1
    data['Luminosity_Index'] = -0.13
    data['SigmoidOfAreas'] = 0.5
    
    df = pd.DataFrame(data, index=[0])
    return df

input_df = user_input_features()

def preprocess_input(df):
    df = df.copy()
    
    df['X_Range'] = df['X_Maximum'] - df['X_Minimum']
    df['Y_Range'] = df['Y_Maximum'] - df['Y_Minimum']
    
    x_range = df['X_Range'].replace(0, 1)
    y_range = df['Y_Range'].replace(0, 1)
    
    df['Density'] = df['Pixels_Areas'] / (x_range * y_range)
    df['Aspect_Ratio'] = df['X_Range'] / y_range
    df['Luminosity_Range'] = df['Maximum_of_Luminosity'] - df['Minimum_of_Luminosity']
    
    skewed_features = ['Pixels_Areas', 'Sum_of_Luminosity', 
                       'X_Range', 'Y_Range', 'Aspect_Ratio']
        
    for col in feature_names:
        if col not in df.columns:
            df[col] = 0 
            
    for feature in skewed_features:
        if feature in df.columns:
            df[feature] = np.log1p(df[feature].abs())
            
    df = df[feature_names]
    return df

if st.button('🔎 Analyze Steel Plate'):
    processed_df = preprocess_input(input_df)
    scaled_array = scaler.transform(processed_df)
    scaled_df = pd.DataFrame(scaled_array, columns=feature_names)
    results = {}
    for target in targets:
        model = models[target]
        prob = model.predict_proba(scaled_df)[0][1]
        results[target] = prob
    st.subheader("Analysis Results")
    results_df = pd.DataFrame(list(results.items()), columns=['Defect Type', 'Probability'])
    results_df = results_df.sort_values(by='Probability', ascending=False)
    top_defect = results_df.iloc[0]
    if top_defect['Probability'] > 0.5:
        st.error(f"⚠️ High Risk Detected: **{top_defect['Defect Type']}** ({top_defect['Probability']:.1%})")
    else:
        st.success("✅ No severe defects detected (All probabilities < 50%)")
    st.bar_chart(results_df.set_index('Defect Type'))
    with st.expander("See Detailed Probabilities"):
        st.dataframe(results_df.style.format({'Probability': '{:.2%}'}))