import gradio as gr
import tensorflow as tf
import numpy as np
from PIL import Image

# Inisialisasi model secara global sebagai None
model = None

classes = ["Abyssinian", "Bengal", "Bombay", "British Shorthair", "Domestic",
           "Maine Coon", "Persian", "Ragdoll", "Siamese", "Sphynx"]

def predict_cat(image):
    global model
    if image is None:
        return "Error", 0.0
    
    # Memuat model secara Lazy Loading di dalam fungsi agar aman saat startup
    if model is None:
        model = tf.keras.models.load_model('cat_model.keras', compile=False)
    
    if image.mode != 'RGB':
        image = image.convert('RGB')
        
    img = image.resize((224, 224))
    img_array = np.array(img) / 255.0
    img_array = np.expand_dims(img_array, axis=0)
    
    # Inferensi dijalankan dengan aman
    prediction = model.predict(img_array)
    predicted_class = np.argmax(prediction, axis=1)[0]
    confidence_score = float(np.max(prediction) * 100)
    
    breed = classes[predicted_class]
    return breed, confidence_score

# Pastikan nama variabel demo menggunakan 'demo' alih-alih 'iface'
demo = gr.Interface(
    fn=predict_cat,
    inputs=gr.Image(type="pil"),
    outputs=[gr.Textbox(label="Breed"), gr.Number(label="Confidence")],
    title="Cat Breed Classifier"
)

# Tambahkan parameter server_name dan port secara eksplisit
demo.launch(server_name="0.0.0.0", server_port=7860)