import gradio as gr
import numpy as np
import tensorflow as tf
import joblib
import os
from huggingface_hub import InferenceClient

# === Load model & scaler ===
lstm_model = tf.keras.models.load_model('retrained_lstm_model.h5', compile=False)
lstm_model.compile(optimizer='adam', loss='mse', metrics=['mae'])

scaler_X = joblib.load('scaler_X.joblib')
scaler_y = joblib.load('scaler_y.joblib')
window_size = 7

# === Setup Inference Client for Gemma 2-27B ===
api_key = os.environ.get("hf_")  # Pastikan HF_TOKEN diset di Secrets
client = InferenceClient(
    provider="nebius",
    api_key=api_key
)

def interpret_rainfall_with_gemma2(tavg, rh_avg, rainfall_mm):
    prompt = f"""
    Data cuaca:
    - Temperatur Rata-rata: {tavg}°C
    - Kelembaban Rata-rata: {rh_avg}%
    - Prediksi Curah Hujan: {rainfall_mm:.2f} mm

    Berikan analisis cuaca berdasarkan data di atas. Apakah hujan tergolong ringan, sedang, atau lebat?
    Apa rekomendasi aktivitas harian, terutama jika saya ingin keluar rumah?
    """

    try:
        completion = client.chat.completions.create(
            model="google/gemma-2-27b-it",
            messages=[{"role": "user", "content": prompt}],
            max_tokens=300
        )
        return completion.choices[0].message.content
    except Exception as e:
        return f"Gagal mendapatkan interpretasi: {str(e)}"

# === Gradio App ===
with gr.Blocks(fill_height=True, title="Prediksi Curah Hujan & Interpretasi AI") as demo:
    gr.Markdown("# Prediksi Curah Hujan dengan LSTM + Interpretasi Gemma 2")
    gr.Markdown("Masukkan data cuaca harian dan dapatkan prediksi serta analisis bahasa alami.")

    with gr.Row():
        with gr.Column():
            tavg_input = gr.Number(label="Temperatur Rata-rata (°C)", value=25.0)
            rh_avg_input = gr.Number(label="Kelembaban Rata-rata (%)", value=80.0)
            ss_input = gr.Number(label="Durasi Sinar Matahari (jam)", value=5.0)
            predict_btn = gr.Button("Prediksi Curah Hujan")

        with gr.Column():
            rainfall_output = gr.Textbox(label="Hasil Prediksi (mm)")
            interpretation_output = gr.Textbox(label="Interpretasi AI", lines=12)

    def predict_and_interpret(tavg, rh_avg, ss):
        user_input = np.array([[tavg, rh_avg, ss]] * window_size)
        user_input_scaled = scaler_X.transform(user_input)
        user_input_scaled = user_input_scaled.reshape(1, window_size, 3)

        predicted_scaled = lstm_model.predict(user_input_scaled)
        predicted_rainfall = scaler_y.inverse_transform(predicted_scaled)
        rainfall_mm = float(predicted_rainfall[0][0])

        explanation = interpret_rainfall_with_gemma2(tavg, rh_avg, rainfall_mm)

        return f"{rainfall_mm:.2f}", explanation

    predict_btn.click(
        predict_and_interpret,
        inputs=[tavg_input, rh_avg_input, ss_input],
        outputs=[rainfall_output, interpretation_output]
    )

    gr.Markdown("### Tentang Aplikasi")
    gr.Markdown("""
    Aplikasi ini memanfaatkan:
    - Model LSTM untuk prediksi curah hujan berdasarkan data suhu, kelembaban, dan durasi sinar matahari
    - Model bahasa besar Gemma 2 (27B) untuk menjelaskan hasil prediksi
    """)

demo.launch()