# app.py
import gradio as gr
import matplotlib.pyplot as plt
import pandas as pd
import joblib
import traceback

# ------------------------------
# Helper: safe load joblib
# ------------------------------
def safe_load(path, name):
    try:
        obj = joblib.load(path)
        print(f"✅ {name} loaded from {path}")
        return obj
    except Exception as e:
        print(f"❌ Error loading {name}: {e}")
        raise

# ------------------------------
# LOAD MODELS & PREPROCESSOR
# ------------------------------
print("Loading models...")
preprocessor = safe_load("preprocessor.pkl", "Preprocessor")
lr_model = safe_load("lr_model.pkl", "Linear Regression")
dt_model = safe_load("dt_model.pkl", "Decision Tree")
rf_model = safe_load("rf_model.pkl", "Random Forest")

loaded_models = {
    "Linear Regression": lr_model,
    "Decision Tree": dt_model,
    "Random Forest": rf_model
}

# ------------------------------
# LOAD DATASET BENCHMARK
# ------------------------------
try:
    df_raw = pd.read_csv("job_salary_mean.csv")
    df_benchmark = df_raw.rename(columns={
        "Judul Pekerjaan": "judul",
        "Perusahaan": "perusahaan",
        "Lokasi": "lokasi",
        "Gaji_Rata2": "gaji"
    })
    df_benchmark["judul_clean"] = df_benchmark["judul"].astype(str).str.lower()
    df_benchmark["lokasi_clean"] = df_benchmark["lokasi"].astype(str).str.lower()
    df_benchmark = df_benchmark.dropna(subset=["judul_clean", "lokasi_clean", "gaji"])
    print(f"✅ Benchmark loaded: {len(df_benchmark)} rows")
except:
    print("❌ job_salary_mean.csv not found")
    df_benchmark = pd.DataFrame(columns=["judul_clean", "lokasi_clean", "gaji"])

# ------------------------------
# LOAD WILAYAH
# ------------------------------
try:
    geo = pd.read_csv("dataset kabupaten indonesia.csv")
    geo = geo[["name", "Unnamed: 3"]].rename(columns={
        "name": "kota",
        "Unnamed: 3": "provinsi"
    })
    geo["kota_clean"] = geo["kota"].astype(str).str.lower().str.replace("kota ", "").replace("kabupaten ", "")
    geo["provinsi"] = geo["provinsi"].astype(str).str.upper()
    MASTER_WILAYAH = pd.Series(geo.provinsi.values, index=geo.kota_clean).to_dict()
    print(f"✅ Loaded {len(MASTER_WILAYAH)} wilayah")
except:
    print("⚠ dataset kabupaten indonesia.csv tidak ada")
    MASTER_WILAYAH = {}

# ------------------------------
# WILAYAH FUNCTIONS
# ------------------------------
def get_pulau_from_provinsi(p):
    p = p.upper()
    if any(x in p for x in ["JAWA", "DKI", "BANTEN"]): return "PULAU JAWA"
    if any(x in p for x in ["SUMATERA", "ACEH", "RIAU"]): return "PULAU SUMATERA"
    if "KALIMANTAN" in p: return "PULAU KALIMANTAN"
    if "SULAWESI" in p: return "PULAU SULAWESI"
    if any(x in p for x in ["BALI", "NUSA"]): return "BALI & NUSA TENGGARA"
    if any(x in p for x in ["PAPUA", "MALUKU"]): return "PAPUA & MALUKU"
    return "INDONESIA"

def deteksi_wilayah(text):
    txt = str(text).lower()
    for kota, prov in MASTER_WILAYAH.items():
        if kota in txt:
            return prov, get_pulau_from_provinsi(prov)
    return "INDONESIA", "INDONESIA"

# ------------------------------
# PREDIKSI + BENCHMARK
# ------------------------------
def analisis_gaji_final(judul, lokasi, model_choice):
    try:
        if not judul or not lokasi:
            return "<b style='color:red;'>Mohon masukkan posisi dan lokasi.</b>", None

        model = loaded_models.get(model_choice)

        df_input = pd.DataFrame({
            "judul_clean": [judul.lower()],
            "lokasi_clean": [lokasi.lower()],
            "perusahaan": ["unknown"]
        })

        try:
            pred = float(model.predict(df_input)[0])
            pred = max(0, pred)
        except Exception as e:
            return f"<b>Gagal memprediksi:</b> {e}", None

        # Benchmark job
        job_match = df_benchmark[df_benchmark["judul_clean"].str.contains(judul.lower(), na=False)]
        max_job = float(job_match["gaji"].max()) if not job_match.empty else pred * 1.3

        # Benchmark location
        provinsi, pulau = deteksi_wilayah(lokasi)
        region_match = df_benchmark[df_benchmark["lokasi_clean"].str.contains(pulau.split()[-1].lower(), na=False)]
        max_reg = float(region_match["gaji"].max()) if not region_match.empty else pred * 1.6

        # Graph
        fig, ax = plt.subplots(figsize=(8,4))
        labels = ["Prediksi Anda", "Max Nasional", "Max Regional"]
        values = [pred, max_job, max_reg]
        ax.bar(labels, values)
        ax.set_title(f"Analisis Gaji: {judul} ({provinsi})")
        ax.set_ylabel("Rp")

        # HTML clean
        html = f"""
        <div style='padding:14px; border-radius:10px; background:#f8fafc'>
            <h3>💰 Estimasi Gaji: Rp {pred:,.0f}</h3>
            <p>📍 Lokasi terdeteksi: <b>{provinsi}</b> — {pulau}</p>
            <p>Max Nasional posisi ini: <b>Rp {max_job:,.0f}</b></p>
            <p>Max Regional: <b>Rp {max_reg:,.0f}</b></p>
        </div>
        """

        return html, fig

    except Exception as e:
        return f"<b>Error:</b> {e}", None

import gradio as gr
# Pastikan Anda memiliki variabel loaded_models dan fungsi analisis_gaji_final yang terdefinisi

# -------------------------------------------
# ENHANCED DEFAULT GRADIO UI (CLEAN & MODERN)
# -------------------------------------------

custom_css = """
<style>

/* Maintain original Gradio feel */
.gradio-container { 
    max-width: 1150px !important; 
    margin: auto; 
    padding-top: 20px;
    font-family: 'Inter', sans-serif;
}

/* Subtle card styling */
.dashboard-box {
    background: white;
    padding: 18px 20px;
    border-radius: 12px;
    border: 1px solid #e5e7eb;
    box-shadow: 0px 1px 4px rgba(0,0,0,0.05);
}

/* Header */
h1 {
    font-weight: 700;
    margin-bottom: 6px;
    font-size: 32px;
}
h3 {
    font-weight: 600;
    margin-bottom: 12px;
}

/* Buttons — keep default look but nicer */
button.primary {
    background: #3b82f6 !important; /* Gradio blue, enhanced */
    color: white !important;
    border-radius: 8px !important;
    padding: 10px 14px !important;
    font-size: 15px !important;
}

/* Inputs enhance */
input, textarea, select, .gr-text-input {
    border-radius: 8px !important;
}

/* Row spacing */
.gr-row {
    gap: 16px !important;
}

/* Responsiveness */
@media (max-width: 768px) {
    .gradio-container {max-width: 95% !important;}
    h1 {font-size: 26px;}
}
</style>
"""

with gr.Blocks(title="Salary AI") as demo:

    # -------------------------------------------
    # Inject CSS
    # -------------------------------------------
    gr.HTML(custom_css)

    # Header
    gr.Markdown("""
    <h1 style='text-align:center;'>💼 Salary AI Dashboard</h1>
    <p style='text-align:center; color:#555; font-size:15px;'>
        Prediksi gaji dengan tampilan simple & elegan ala Gradio.
    </p>
    """)

    with gr.Row():

        # LEFT — Input Form
        with gr.Column(scale=1, min_width=360):
            gr.HTML("<div class='dashboard-box'><h3>📥 Input</h3>")
            t1 = gr.Textbox(label="Posisi Pekerjaan", placeholder="cth: Data Analyst")
            t2 = gr.Textbox(label="Kabupaten/Kota", placeholder="cth: Bandung")
            model = gr.Dropdown(
                choices=list(loaded_models.keys()),
                value="Random Forest",
                label="Model Prediksi"
            )
            btn = gr.Button("🔍 Analisis Gaji", variant="primary")
            gr.HTML("</div>")

        # RIGHT — Output
        with gr.Column(scale=2):
            gr.HTML("<div class='dashboard-box'><h3>📊 Hasil</h3>")
            out_html = gr.HTML()
            out_plot = gr.Plot()
            gr.HTML("</div>")

    # Button Function
    btn.click(
        analisis_gaji_final,
        inputs=[t1, t2, model],
        outputs=[out_html, out_plot]
    )

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