app.py

import streamlit as st
import pandas as pd
import joblib
import shap
import matplotlib.pyplot as plt

# Load your trained pipeline and label encoder
pipeline = joblib.load('xgb_pipeline.joblib')  # your Pipeline object
le = joblib.load('label_encoder.joblib')       # your LabelEncoder

st.title('Cheese Milk Type Classifier with SHAP Explanation')

# ===== User Input =====
country = st.selectbox('Country', [
    'France', 'Italy', 'Switzerland', 'United States', 'Germany',
    'Spain', 'England', 'Australia', 'Canada', 'Greece'
])

texture = st.selectbox('Texture', [
    'creamy', 'firm', 'crumbly', 'smooth', 'hard', 'soft', 'dense',
    'elastic', 'dry', 'buttery', 'mild', 'open'
])

rind = st.selectbox('Rind', [
    'natural', 'washed', 'none', 'ash coated', 'leaf wrapped', 'Unknown'
])

vegetarian = st.selectbox('Vegetarian Status', ['True', 'False'])

flavor = st.selectbox('Flavor', [
    'nutty', 'mild', 'sweet', 'earthy', 'fruity', 'acidic', 'buttery',
    'smokey', 'full-flavored', 'salty', 'creamy', 'sharp', 'bitter', 'Unknown'
])

input_df = pd.DataFrame({
    'country': [country],
    'texture': [texture],
    'rind': [rind],
    'vegetarian': [vegetarian],
    'flavor': [flavor]
})

# ===== Prediction and SHAP =====
if st.button('Predict', key='predict_button'):
    # 1️⃣ Make prediction
    pred_encoded = pipeline.predict(input_df)[0]
    pred_label = le.inverse_transform([pred_encoded])[0]
    st.write(f'Predicted Milk Type: **{pred_label}**')

    # 2️⃣ SHAP Explanation
    # Extract XGBoost model from pipeline
    model = pipeline.named_steps['clf']  # 'clf' is your XGBClassifier step

    # Transform input (excluding the model)
    input_transformed = pipeline[:-1].transform(input_df)

    # Create SHAP explainer
    explainer = shap.Explainer(model)
    shap_values = explainer(input_transformed)

    # Plot SHAP values (bar plot for top features)
    st.subheader('Feature Contribution (SHAP Values)')
    fig, ax = plt.subplots()
    shap.plots.bar(shap_values[0], max_display=10, show=False)  # top 10 features
    st.pyplot(fig)