app.py

import gradio as gr
from sentence_transformers import SentenceTransformer
from sentence_transformers.util import cos_sim
import torch
import logging

# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Global model variable
model = None

def load_model():
    """Load the sentence transformer model"""
    global model
    if model is None:
        try:
            logger.info("Loading sentence transformer model...")
            model = SentenceTransformer('nabil-tazi/autotrain-d19rl-a8u4f')
            logger.info("Model loaded successfully!")
        except Exception as e:
            logger.error(f"Failed to load model: {e}")
            raise e
    return model

def classify_ambiance(user_input):
    """Classify lighting ambiance from user input"""
    
    if not user_input or not user_input.strip():
        return "❌ Please enter some text", {}, ""
    
    try:
        # Load model if not already loaded
        current_model = load_model()
        
        # Your three reference ambiances
        references = ["bright", "cozy", "natural"]
        
        # Get embeddings
        user_embedding = current_model.encode([user_input.strip()])
        ref_embeddings = current_model.encode(references)
        
        # Calculate similarities
        similarities = cos_sim(user_embedding, ref_embeddings)[0]
        
        # Get best match
        best_idx = similarities.argmax()
        best_ambiance = references[best_idx]
        confidence = float(similarities[best_idx])
        
        # Format all scores for debugging
        all_scores = {ref: round(float(sim), 4) for ref, sim in zip(references, similarities)}
        
        # Create result with emoji
        emoji_map = {"bright": "☀️", "cozy": "🕯️", "natural": "🌿"}
        result_text = f"## {emoji_map.get(best_ambiance, '💡')} **{best_ambiance.upper()}**\n**Confidence:** {confidence:.3f}"
        
        # Create confidence bar
        confidence_bar = f"**Confidence Level:** {'█' * int(confidence * 20)}{'░' * (20 - int(confidence * 20))} {confidence:.1%}"
        
        logger.info(f"Classified '{user_input}' as '{best_ambiance}' with confidence {confidence:.3f}")
        
        return result_text, all_scores, confidence_bar
        
    except Exception as e:
        error_msg = f"❌ Error: {str(e)}"
        logger.error(f"Classification error: {e}")
        return error_msg, {}, ""

# Create Gradio interface
with gr.Blocks(
    title="🏠 Lighting Ambiance Classifier",
    theme=gr.themes.Soft(),
    css="""
    .gradio-container {
        max-width: 800px !important;
        margin: auto !important;
    }
    .result-box {
        background: linear-gradient(45deg, #f0f0f0, #ffffff);
        border-radius: 10px;
        padding: 20px;
    }
    """
) as demo:
    
    # Header
    gr.Markdown(
        """
        # 🏠 Lighting Ambiance Classifier
        
        **Classify your lighting preferences into three categories:**
        - ☀️ **Bright**: Well-lit, luminous, clear lighting
        - 🕯️ **Cozy**: Warm, dim, soft, ambient lighting  
        - 🌿 **Natural**: Daylight, sunlight, organic lighting
        
        **Supports both English and Japanese!** 🇺🇸🇯🇵
        """
    )
    
    with gr.Row():
        with gr.Column(scale=2):
            # Input section
            gr.Markdown("### 💬 Enter your lighting preference:")
            input_text = gr.Textbox(
                label="Your lighting preference",
                placeholder="e.g., 'not bright', '明るくない', 'cozy lighting', '自然な光が欲しい'",
                lines=3,
                max_lines=5
            )
            
            with gr.Row():
                submit_btn = gr.Button("🔍 Classify", variant="primary", size="lg")
                clear_btn = gr.Button("🗑️ Clear", variant="secondary")
            
        with gr.Column(scale=2):
            # Output section
            gr.Markdown("### 🎯 Classification Result:")
            result = gr.Markdown(value="Enter text and click classify!", elem_classes=["result-box"])
            confidence_bar = gr.Markdown(value="")
    
    # Detailed scores
    with gr.Row():
        scores = gr.JSON(label="📊 Detailed Similarity Scores", visible=True)
    
    # Example inputs
    gr.Markdown("### 💡 Try these examples:")
    with gr.Row():
        examples = gr.Examples(
            examples=[
                ["not bright"],
                ["明るくない"], 
                ["I want cozy lighting"],
                ["自然な光が欲しい"],
                ["make it brighter"],
                ["暗くしたい"],
                ["romantic atmosphere"],
                ["作業しやすい明るさ"],
                ["candle light"],
                ["太陽光みたい"],
                ["harsh fluorescent"],
                ["優しい照明"]
            ],
            inputs=input_text,
            examples_per_page=6
        )
    
    # Footer
    gr.Markdown(
        """
        ---
        **Model:** Fine-tuned multilingual sentence transformer trained on English-Japanese lighting preference pairs.
        
        **How it works:** The model compares your input text with the three ambiance categories and returns the most similar one with a confidence score.
        """
    )
    
    # Event handlers
    def clear_all():
        return "", "Enter text and click classify!", {}, ""
    
    submit_btn.click(
        fn=classify_ambiance,
        inputs=input_text,
        outputs=[result, scores, confidence_bar]
    )
    
    input_text.submit(
        fn=classify_ambiance,
        inputs=input_text,
        outputs=[result, scores, confidence_bar]
    )
    
    clear_btn.click(
        fn=clear_all,
        outputs=[input_text, result, scores, confidence_bar]
    )

# Launch the app
if __name__ == "__main__":
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        show_error=True
    )