import gradio as gr
import numpy as np
import cv2 as cv
import requests
import time
import os

host = os.environ.get("host")
code = os.environ.get("code")
model_llm = os.environ.get("model")
content = os.environ.get("content")
state = os.environ.get("state")
data = None
model = None
image = None
prediction = None
labels = None

print('START')
np.set_printoptions(suppress=True)

data = np.ndarray(shape=(1, 224, 224, 3), dtype=np.float32)

with open("labels.txt", "r") as file:
    labels = file.read().splitlines()

def classify(image_path, text_input):
    if text_input == code:
        output = []  # Create an empty list for output
        image_data = np.array(image_path)
        image_data = cv.resize(image_data, (224, 224))
        image_array = np.asarray(image_data)
        normalized_image_array = (image_array.astype(np.float32) / 127.0) - 1
        data[0] = normalized_image_array
    
        # Load the model within the classify function
        import tensorflow as tf
        model = tf.keras.models.load_model('keras_model.h5')
    
        prediction = model.predict(data)
        
        max_label_index = None
        max_prediction_value = -1

        print('Prediction')
    
        for i, label in enumerate(labels):
            prediction_value = float(prediction[0][i])
            rounded_value = round(prediction_value, 2)
            print(f'{label}: {rounded_value}')
    
            if prediction_value > max_prediction_value:
                max_label_index = i
                max_prediction_value = prediction_value  # Update max_prediction_value
    
        if max_label_index is not None:
            max_label = labels[max_label_index].split(' ', 1)[1]
            print(f'Maximum Prediction: {max_label} with a value of {round(max_prediction_value, 2)}')
    
            time.sleep(1)
            print("\nWays to dispose of this waste: " + max_label)
            payload = [
                {"role": "system", "content": "You are a helpful assistant."},
                {"role": "user", "content": "Give me the steps to dispose of this waste in bullet points (5 max): " + max_label}
            ]
    
            response = requests.post(host, json={
                "messages": payload,
                "model": model_llm,
                "temperature": 0.5,
                "presence_penalty": 0,
                "frequency_penalty": 0,
                "top_p": 1
            }).json()

            reply = response["choices"][0]["message"]["content"]

            output.append({"type": max_label, "prediction_value": round(max_prediction_value, 2), "content": reply})
        
        return output  # Return the populated output list
    else:
        return "Unauthorized"

iface = gr.Interface(
    fn=classify,
    inputs=[gr.inputs.Image(), "text"],
    outputs=gr.outputs.JSON(),  # Output as JSON
    title="Waste Classifier",
    description="Upload an image to classify and get disposal instructions."
)
iface.launch()