# app.py - FINAL POLISHED version with forgiving input clamping

import gradio as gr
import joblib
import pandas as pd
import os
import subprocess

# --- STEP 1: Download Model Files ---
MODEL_URL = "https://huggingface.co/szili2011/ai-house-price-predictor/resolve/main/housing_model.joblib"
COLUMNS_URL = "https://huggingface.co/szili2011/ai-house-price-predictor/resolve/main/model_columns.joblib"
MODEL_LOCAL_PATH = "housing_model.joblib"
COLUMNS_LOCAL_PATH = "model_columns.joblib"

if not os.path.exists(MODEL_LOCAL_PATH):
    print("--- Downloading Model Files ---")
    subprocess.run(["wget", "-O", MODEL_LOCAL_PATH, MODEL_URL])
    subprocess.run(["wget", "-O", COLUMNS_LOCAL_PATH, COLUMNS_URL])
    print("--- Download Complete ---")

# --- STEP 2: Load Model and Columns ---
try:
    model = joblib.load(MODEL_LOCAL_PATH)
    model_columns = joblib.load(COLUMNS_LOCAL_PATH)
    print("✅ Model and columns loaded successfully.")
    model_loaded_successfully = True
except Exception as e:
    print(f"❌ CRITICAL ERROR during model loading: {e}")
    model_loaded_successfully = False


# --- Core prediction function with the new, forgiving clamping logic ---
def predict_price(sqft, bedrooms, house_age, condition, year_sold, interest_rate, region, sub_type, style, has_garage, has_pool):
    if not model_loaded_successfully:
        raise gr.Error("Model is not loaded. Please check the Space logs for errors.")

    # --- NEW: Forgiving Input Clamping ---
    # Instead of raising an error, we gently guide extreme inputs back into a reasonable range.
    # The min() function takes the smaller of two numbers, the max() function takes the larger.
    # This "clamps" the value between a minimum and a maximum.
    
    # Clamp square footage between 300 and 30,000
    sqft_clamped = max(300, min(sqft, 30000))
    
    # Clamp bedrooms between 1 and 20
    bedrooms_clamped = max(1, min(bedrooms, 20))
    
    # Clamp house age between 0 and 200
    house_age_clamped = max(0, min(house_age, 200))

    # Clamp year sold between 2000 and 2030
    year_sold_clamped = max(2000, min(year_sold, 2030))

    # Clamp interest rate between 0 and 25
    interest_rate_clamped = max(0.0, min(interest_rate, 25.0))

    # We now use these "safe" clamped values for the prediction.
    input_data = {
        'SquareFootage': sqft_clamped,
        'Bedrooms': bedrooms_clamped,
        'HouseAge': house_age_clamped,
        'PropertyCondition': condition,
        'HasGarage': has_garage,
        'HasPool': has_pool,
        'YearSold': year_sold_clamped,
        'InterestRate': interest_rate_clamped,
        'Region': region,
        'SubType': sub_type,
        'ArchitecturalStyle': style
    }
    input_df = pd.DataFrame([input_data])
    
    input_processed = pd.get_dummies(input_df)
    final_input = input_processed.reindex(columns=model_columns, fill_value=0)
    
    predicted_price = model.predict(final_input)[0]
    
    return f"${predicted_price:,.0f}"

# --- Gradio Interface with UI limits ---
# The UI limits still provide a good first line of guidance for the user.
demo = gr.Interface(
    fn=predict_price,
    inputs=[
        gr.Number(label="Square Footage", value=2500, minimum=300, maximum=30000),
        gr.Number(label="Bedrooms", value=4, minimum=1, maximum=20),
        gr.Number(label="House Age (years)", value=15, minimum=0, maximum=200),
        gr.Slider(label="Property Condition", minimum=1, maximum=10, step=1, value=8),
        gr.Number(label="Year Sold", value=2024, minimum=2000, maximum=2030),
        gr.Number(label="Interest Rate (%)", value=5.5, minimum=0, maximum=25),
        gr.Radio(['Sunbelt', 'Pacific Northwest', 'Rust Belt', 'New England', 'Mountain West'], label="Region", value="Sunbelt"),
        gr.Radio(['Urban', 'Suburban', 'Rural', 'Historic District'], label="Sub-Type", value="Suburban"),
        gr.Radio(['Modern', 'Ranch', 'Colonial', 'Craftsman', 'Victorian'], label="Architectural Style", value="Colonial"),
        gr.Checkbox(label="Has Garage?", value=True),
        gr.Checkbox(label="Has Pool?", value=False)
    ],
    outputs=gr.Textbox(label="Predicted Price"),
    title="AI House Price Predictor",
    description="Describe a property, and our AI will estimate its market value. The model is robust and will provide estimates even for extreme values by capping them to its known limits."
)

# Launch the app
demo.launch()