app.py

import pickle
import pandas as pd
import shap
import gradio as gr
import matplotlib.pyplot as plt

# Load model + SHAP
model = pickle.load(open("salar_xgb_team.pkl", "rb"))
explainer = shap.Explainer(model)

# Education map
education_map = {
    "Preschool": 1, "1st-4th": 2, "5th-6th": 3, "7th-8th": 4, "9th": 5,
    "10th": 6, "11th": 7, "12th": 8, "HS-grad": 9, "Some-college": 10,
    "Assoc-voc": 11, "Assoc-acdm": 12, "Bachelors": 13, "Masters": 14,
    "Prof-school": 15, "Doctorate": 16
}

def predict_user_profile(age, education_label, sex, capital_gain, capital_loss, hours_per_week):
    education_num = education_map.get(education_label, 9)
    sex_binary = 0 if sex == "Male" else 1

    df = pd.DataFrame([{
        "age": age,
        "education-num": education_num,
        "sex": sex_binary,
        "capital-gain": capital_gain,
        "capital-loss": capital_loss,
        "hours-per-week": hours_per_week
    }])

    prob = model.predict_proba(df)[0]
    shap_values = explainer(df)
    predicted_class = int(prob[1] > 0.5)
    confidence = prob[1] if predicted_class else prob[0]

    # Confidence bar
    level = int(confidence * 10)
    bar = "🟩" * level + "⬜" * (10 - level)

    # Short user story
    story = (
        f"📘 **Meet your profile**: A {age}-year-old {'woman' if sex_binary else 'man'} "
        f"with {education_label} education, working {hours_per_week} hrs/week "
        f"and reporting gains/losses of ${capital_gain}/{capital_loss}."
    )

    # Insight summary
    result = f"🔮 **Prediction**: {'> $50K' if predicted_class else '≤ $50K'}\n"
    result += f"📊 **Confidence**: {confidence:.2%}\n\n"
    result += f"{bar} Confidence Meter\n\n"
    result += f"{story}"

    # Explanation from SHAP
    sorted_features = sorted(zip(df.columns, shap_values.values[0]), key=lambda x: abs(x[1]), reverse=True)
    top_feature = sorted_features[0][0].replace('-', ' ').capitalize()
    direction = "increased" if sorted_features[0][1] > 0 else "reduced"
    meaning = f"💡 *Your {top_feature} likely {direction} your predicted income.*"

    # Optional: add a tip
    tip = (
        "💬 Tip: Consider exploring upskilling programs for higher income opportunities!" if predicted_class == 0
        else "💬 Tip: Keep optimizing your time and gain strategy for even higher returns!"
    )

    # SHAP plot
    plt.figure(figsize=(6, 3))
    shap.plots.bar(shap_values[0], max_display=6, show=False)
    plt.tight_layout()
    fig = plt.gcf()
    plt.close()

    return {"≤ $50K": round(prob[0], 2), "> $50K": round(prob[1], 2)}, fig, f"{result}\n\n{meaning}\n\n{tip}"

# App layout
with gr.Blocks(title="🧠 Group 5 | Income Story Predictor") as demo:
    gr.Markdown("## 🧠 Group 5 | Income Story Predictor")
    gr.Markdown("Welcome to your **Income Story** — enter your details and discover your predicted earnings path, confidence level, and key factors influencing it.")

    with gr.Row():
        with gr.Column(scale=1):
            summary = gr.Textbox(label="Prediction Summary", lines=8, interactive=False)
            result_label = gr.Label(label="Class Probabilities")
            shap_plot = gr.Plot(label="Key Feature Contributions")

        with gr.Column(scale=1):
            age = gr.Slider(18, 100, step=1, label="🎂 Age", value=35)
            education_label = gr.Dropdown(list(education_map.keys()), label="🎓 Education Level", value="HS-grad")
            sex = gr.Radio(["Male", "Female"], label="🧍 Sex", value="Male")
            capital_gain = gr.Number(label="📈 Capital Gain", value=0)
            capital_loss = gr.Number(label="📉 Capital Loss", value=0)
            hours_per_week = gr.Slider(1, 100, step=1, label="⌛ Hours per Week", value=40)
            analyze = gr.Button("✨ Tell My Income Story")

    analyze.click(
        fn=predict_user_profile,
        inputs=[age, education_label, sex, capital_gain, capital_loss, hours_per_week],
        outputs=[result_label, shap_plot, summary]
    )

    gr.Markdown("---")
    gr.Markdown("Group 5 :)")

demo.launch()