from pathlib import Path
import pickle

import gradio as gr
import pandas as pd


BASE_DIR = Path(__file__).resolve().parent
MODEL_PATH = BASE_DIR / "apartment_price_model.pkl"
DATA_PATH = BASE_DIR / "original_apartment_data_analytics_hs24.csv"


with MODEL_PATH.open("rb") as f:
    model = pickle.load(f)


df = pd.read_csv(DATA_PATH)
municipality_columns = ["town", "postalcode", "pop", "pop_dens", "frg_pct", "emp", "tax_income"]
municipality_df = df[municipality_columns].dropna().copy()
municipality_df["town"] = municipality_df["town"].astype(str).str.strip()
municipality_df = municipality_df.drop_duplicates(subset=["town"]).sort_values("town")
MUNICIPALITY_DATA = municipality_df.set_index("town").to_dict("index")


def _build_features(rooms: float, area: float, town: str) -> tuple[pd.DataFrame, dict]:
    municipality = MUNICIPALITY_DATA[town]

    pop = float(municipality["pop"])
    pop_dens = float(municipality["pop_dens"])
    frg_pct = float(municipality["frg_pct"])
    emp = float(municipality["emp"])
    tax_income = float(municipality["tax_income"])

    municipality_area_proxy = pop / pop_dens
    emp_per_resident = emp / pop
    foreigner_count_est = pop * (frg_pct / 100)

    features = pd.DataFrame(
        [
            {
                "rooms": rooms,
                "area": area,
                "pop": pop,
                "pop_dens": pop_dens,
                "frg_pct": frg_pct,
                "emp": emp,
                "tax_income": tax_income,
                "municipality_area_proxy": municipality_area_proxy,
                "emp_per_resident": emp_per_resident,
                "foreigner_count_est": foreigner_count_est,
            }
        ]
    )

    return features, municipality


def predict_price(rooms: float, area: float, town: str) -> tuple[str, str]:
    features, municipality = _build_features(rooms=rooms, area=area, town=town)
    prediction = float(model.predict(features)[0])

    pop = float(municipality["pop"])
    pop_dens = float(municipality["pop_dens"])
    frg_pct = float(municipality["frg_pct"])
    emp = float(municipality["emp"])
    
    municipality_area_proxy = pop / pop_dens
    emp_per_resident = emp / pop
    foreigner_count_est = pop * (frg_pct / 100)

    details = (
        f"Ort: {town} (PLZ {int(municipality['postalcode'])})\n\n"
        f"Gemeindedaten:\n"
        f"  Bevölkerung: {int(pop):,}\n"
        f"  Bevölkerungsdichte: {pop_dens:.1f}/km²\n"
        f"  Ausländeranteil: {frg_pct:.1f}%\n"
        f"  Beschäftigte: {int(emp):,}\n"
        f"  Steuerbares Einkommen: CHF {municipality['tax_income']:,.0f}\n\n"
        f"Zusätzliche berechnete Features:\n"
        f"  Gemeindegröße: {municipality_area_proxy:.2f} km²\n"
        f"  Arbeitsplatzquote: {emp_per_resident:.3f}\n"
        f"  Ausländerpopulation: {int(foreigner_count_est):,}"
    )

    return f"CHF {prediction:,.2f} pro Monat", details


demo = gr.Interface(
    fn=predict_price,
    inputs=[
        gr.Slider(1, 8, value=3, step=0.5, label="Zimmer"),
        gr.Slider(20, 250, value=80, step=1, label="Wohnfläche (m²)"),
        gr.Dropdown(choices=list(MUNICIPALITY_DATA.keys()), value="Zürich", label="Gemeinde"),
    ],
    outputs=[
        gr.Textbox(label="Geschätzte Miete"),
        gr.Textbox(label="Verwendete Gemeindedaten"),
    ],
    examples=[
        [2.5, 60, "Zürich"],
        [3.5, 90, "Winterthur"],
        [4.5, 120, "Uster"],
    ],
    title="Apartment Price Prediction – Kanton Zürich",
    description=(
        "Vorhersage der monatlichen Wohnungsmiete mit einem Random-Forest-Regressionsmodell. "
        "Die App nutzt Basismerkmale und zusätzliche Feature-Engineering-Variablen "
        "(municipality_area_proxy, emp_per_resident, foreigner_count_est)."
    ),
)


if __name__ == "__main__":
    demo.launch()