app.py

import gradio as gr
import tensorflow as tf
import numpy as np
from tensorflow.keras.layers import TextVectorization


MAX_FEATURES = 200000
OUTPUT_SEQUENCE_LENGTH = 1800

class ToxicityClassifier:
    def __init__(self, model_path='toxicity.h5'):
        """Initialize the toxicity classifier with model and vectorizer."""
        try:
            self.model = tf.keras.models.load_model(model_path)
            self.vectorizer = self._initialize_vectorizer()
            print("✅ Model and vectorizer loaded successfully!")
        except Exception as e:
            raise RuntimeError(f"Failed to initialize classifier: {str(e)}")

    def _initialize_vectorizer(self):
        """Initialize and adapt text vectorizer (dummy adapt for demo)."""
        vectorizer = TextVectorization(
            max_tokens=MAX_FEATURES,
            output_sequence_length=OUTPUT_SEQUENCE_LENGTH,
            output_mode='int'
        )
        vectorizer.adapt(["dummy text"])
        return vectorizer

    def predict_toxicity(self, text):
        """Predict toxicity probabilities for input text."""
        try:
            # Vectorize and predict
            vectorized_text = self.vectorizer([text])
            predictions = self.model.predict(vectorized_text, verbose=0)[0]
            
            # Format results
            results = {
                "toxic": float(predictions[0]),
                "severe_toxic": float(predictions[1]),
                "obscene": float(predictions[2]),
                "threat": float(predictions[3]),
                "insult": float(predictions[4]),
                "identity_hate": float(predictions[5])
            }
            return results
        except Exception as e:
            print(f"Prediction error: {e}")
            return None

def create_interface():
    """Create and configure Gradio interface."""
    classifier = ToxicityClassifier()
    
    def classify_text(text):
        results = classifier.predict_toxicity(text)
        if results is None:
            return "Error processing your request. Please try again."
        
        # Format output with emojis and percentages
        output = []
        for label, score in results.items():
            emoji = "🔴" if score > 0.5 else "🟢"
            percentage = f"{score*100:.1f}%"
            output.append(f"{emoji} {label.replace('_', ' ').title()}: {percentage}")
        
        return "\n".join(output)
    
    # Custom CSS for better UI
    css = """
    .gradio-container { max-width: 600px !important; }
    .label { font-weight: bold; }
    .warning { color: #ff4d4d; }
    """
    
    # Interface components
    with gr.Blocks(css=css, title="Toxicity Classifier",theme = gr.themes.Soft()) as app:
        gr.Markdown("""
        # 🚨 Toxicity Classifier
        Enter text to analyze for toxic content. The model evaluates:
        - Toxicity
        - Severe Toxicity
        - Obscenity
        - Threats
        - Insults
        - Identity Hate
        """)
        
        with gr.Row():
            text_input = gr.Textbox(
                placeholder="Enter text to analyze...",
                label="Input Text",
                lines=3,
                max_lines=6
            )
        
        submit_btn = gr.Button("Analyze", variant="primary")
        
        with gr.Group():
            gr.Markdown("### Results")
            output = gr.Textbox(label="Analysis", interactive=False)
        
        examples = gr.Examples(
            examples=[
                "You're an idiot!",
                "I hate you"
            ],
            inputs=text_input,
            label="Try these examples"
        )
        
        submit_btn.click(
            fn=classify_text,
            inputs=text_input,
            outputs=output
        )
    
    return app

if __name__ == "__main__":
    app = create_interface()
    app.launch(
        server_name="0.0.0.0",
        share=True,
        favicon_path=None
    )