import gradio as gr
import pickle
import numpy as np

# Load the trained classifier and TF-IDF vectorizer
try:
    svm_clf = pickle.load(open('clf.pkl', 'rb'))
    svm_tfidf = pickle.load(open('tfidf.pkl', 'rb'))
except Exception as e:
    raise Exception(f"Error loading model files: {str(e)}")

def predict_text(text):
    # Transform the input text using the TF-IDF vectorizer
    text_transformed = svm_tfidf.transform([text])

    # Get prediction (0.0 for human, 1.0 for AI)
    prediction = svm_clf.predict(text_transformed)[0]

    # Get decision score (distance from hyperplane)
    decision_score = svm_clf.decision_function(text_transformed)[0]

    # Sigmoid scaling to get a confidence-like score (0-100%)
    ai_confidence = 100 * (1 / (1 + np.exp(-decision_score)))
    human_confidence = 100 - ai_confidence

    # Return formatted result
    label = "AI" if prediction == 1.0 else "Human"
    return f"Prediction: {label}\nConfidence: {ai_confidence:.2f}% AI, {human_confidence:.2f}% Human"

# Define the Gradio interface
iface = gr.Interface(
    fn=predict_text,
    inputs=gr.Textbox(lines=5, placeholder="Enter your text here..."),
    outputs="text",
    title="AI vs Human Text Classifier",
    description="Enter text to classify it as AI-generated or Human-written, with confidence scores."
)

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