import gradio as gr
import pandas as pd
import pickle
import os

# Helper function to load pickle files
def load_pickle_file(file_name):
    file_path = os.path.join(os.path.dirname(__file__), file_name)
    try:
        if os.path.exists(file_path):
            with open(file_path, 'rb') as file:
                return pickle.load(file)
        else:
            return f"File {file_name} not found."
    except Exception as e:
        return f"An error occurred while loading {file_name}: {e}"

# Load the pre-trained model and label encoder
model = load_pickle_file('best_model.pkl')
label_encoder = load_pickle_file('label_encoder.pkl')

# Ensure model and label encoder are loaded correctly
if isinstance(model, str) or isinstance(label_encoder, str):
    raise Exception(f"Error loading model or label encoder: {model} | {label_encoder}")

# Define the prediction function
def predict_coffee_type(time_of_day, coffee_strength, sweetness_level, milk_type, coffee_temperature, flavored_coffee, caffeine_tolerance, coffee_bean, coffee_size, dietary_preferences):
    # Input Data
    input_data = pd.DataFrame({
        'Token_0': [time_of_day],
        'Token_1': [coffee_strength],
        'Token_2': [sweetness_level],
        'Token_3': [milk_type],
        'Token_4': [coffee_temperature],
        'Token_5': [flavored_coffee],
        'Token_6': [caffeine_tolerance],
        'Token_7': [coffee_bean],
        'Token_8': [coffee_size],
        'Token_9': [dietary_preferences]
    })

    # One-hot encode the input data (ensure it matches the training data)
    input_encoded = pd.get_dummies(input_data)
    required_columns = model.feature_names_in_  # Ensure that the input has the correct columns
    for col in required_columns:
        if col not in input_encoded.columns:
            input_encoded[col] = 0  # Add missing columns as 0
    
    input_encoded = input_encoded[required_columns]  # Ensure the order of columns matches the training data
    
    # Make prediction
    prediction = model.predict(input_encoded)[0]
    
    # Decode the label
    coffee_type = label_encoder.inverse_transform([prediction])[0]
    
    return f"Recommended Coffee: {coffee_type}"

# Set up Gradio interface
interface = gr.Interface(
    fn=predict_coffee_type,
    inputs=[
        gr.inputs.Dropdown(choices=['morning', 'afternoon', 'evening'], label="Time of Day"),
        gr.inputs.Dropdown(choices=['mild', 'regular', 'strong'], label="Coffee Strength"),
        gr.inputs.Dropdown(choices=['unsweetened', 'lightly sweetened', 'sweet'], label="Sweetness Level"),
        gr.inputs.Dropdown(choices=['none', 'regular', 'skim', 'almond'], label="Milk Type"),
        gr.inputs.Dropdown(choices=['hot', 'iced', 'cold brew'], label="Coffee Temperature"),
        gr.inputs.Dropdown(choices=['yes', 'no'], label="Flavored Coffee"),
        gr.inputs.Dropdown(choices=['low', 'medium', 'high'], label="Caffeine Tolerance"),
        gr.inputs.Dropdown(choices=['Arabica', 'Robusta', 'blend'], label="Coffee Bean"),
        gr.inputs.Dropdown(choices=['small', 'medium', 'large'], label="Coffee Size"),
        gr.inputs.Dropdown(choices=['none', 'vegan', 'lactose-intolerant'], label="Dietary Preferences")
    ],
    outputs="text",
    title="Coffee Type Prediction"
)

# Launch the Gradio app
if __name__ == "__main__":
    interface.launch()