app.py

import gradio as gr
import pandas as pd
import joblib

# Load model
model = joblib.load("shelfy_purchase_rate_v2.joblib")

def predict_conversion(product_name, views, carts, unique_users, avg_price, carts_lag_1, carts_lag_3):
    # Convert to float (Gradio passes strings)
    views = float(views or 0)
    carts = float(carts or 0) 
    unique_users = float(unique_users or 0)
    avg_price = float(avg_price or 0)
    carts_lag_1 = float(carts_lag_1 or 0)
    carts_lag_3 = float(carts_lag_3 or 0)
    
    # Calculate ratios (same as training)
    cart_intent = carts / (views + 1)
    user_intensity = unique_users / (views + 1)
    
    data = pd.DataFrame({
        'views': [views],
        'carts': [carts],
        'unique_users': [unique_users],
        'avg_price': [avg_price],
        'carts_lag_1': [carts_lag_1],
        'carts_lag_3': [carts_lag_3],
        'cart_intent': [cart_intent],
        'user_intensity': [user_intensity]
    })
    
    future_rate = model.predict(data)[0]
    
    return (
        f"{future_rate:.2%}", 
        f"${future_rate * views * avg_price:.0f}",
        "Good Demand" if future_rate >= 0.025 else "Medium Demand" if future_rate >= 0.015 else "Low Demand"
    )

demo = gr.Interface(
    fn=predict_conversion,
    inputs=[
        gr.Textbox("iPhone Case", label="Product Name"),
        gr.Number(1500, label="Views (Today)"), 
        gr.Number(45, label="Carts (Today)"),
        gr.Number(320, label="Unique Users"),
        gr.Number(26, label="Avg Price $"),
        gr.Number(40, label="Yesterday Carts"), 
        gr.Number(35, label="3 Days Ago Carts")
    ],
    outputs=[
        gr.Label(label="Predicted Conversion (Tomorrow)"),
        gr.Label(label="Revenue Potential"), 
        gr.Label(label="Demand")
    ],
    title="🛒 Shelfy: Tomorrow's Best Sellers",
    description="Top 10% predictions captured 25.4% of sales!"
)

demo.launch()