import gradio as gr
from transformers import pipeline

# --- Model Loading ---
# Using the model ID you've been working with, which is fine-tuned on an English dataset (IMDB).
# It's crucial that this model is indeed fine-tuned for sentiment analysis in English.
MODEL_ID = "Light-Dav/sentiment-analysis-full-project"

try:
    # Attempt to load the pipeline. This needs to be outside the function for efficiency.
    # The 'return_all_scores=True' is important for getting scores for all labels.
    sentiment_analyzer = pipeline("sentiment-analysis", model=MODEL_ID, return_all_scores=True)
    model_loaded_successfully = True
except Exception as e:
    print(f"Error loading model: {e}")
    sentiment_analyzer = None
    model_loaded_successfully = False

# --- Custom CSS for a unique look ---
custom_css = """
body {
    background-color: #f0f2f5; /* Light grey background */
}
.gradio-container {
    box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1);
    border-radius: 15px;
    overflow: hidden;
    background-color: #ffffff; /* White card background */
}
h1, h2, h3 {
    color: #4CAF50; /* Green accents */
    text-align: center;
    font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
    animation: fadeIn 1s ease-in-out;
}
.gr-button.primary {
    background-color: #4CAF50 !important;
    color: white !important;
    border-radius: 8px;
    transition: background-color 0.3s ease;
}
.gr-button.primary:hover {
    background-color: #45a049 !important;
}
.gradio-output {
    border: 1px solid #e0e0e0;
    border-radius: 10px;
    padding: 15px;
    margin-top: 20px;
    background-color: #f9f9f9;
}
.sentiment-display {
    text-align: center;
    padding: 10px;
    border-radius: 8px;
    margin-top: 15px;
    font-size: 1.2em;
    font-weight: bold;
}
.sentiment-positive {
    background-color: #e6ffe6; /* Light green */
    color: #28a745; /* Darker green */
}
.sentiment-negative {
    background-color: #ffe6e6; /* Light red */
    color: #dc3545; /* Darker red */
}
.sentiment-neutral {
    background-color: #e6f7ff; /* Light blue */
    color: #007bff; /* Darker blue */
}
@keyframes fadeIn {
    from { opacity: 0; }
    to { opacity: 1; }
}
"""

# --- Helper Function for Sentiment Interpretation ---
def interpret_sentiment(label, score):
    emoji = ""
    description = ""
    color_class = ""

    if label.lower() == "positive": # Your model might output 'LABEL_2' or 'POS'
        emoji = "😊"
        description = "This text expresses a **highly positive** sentiment." if score > 0.9 else "This text expresses a **positive** sentiment."
        color_class = "sentiment-positive"
    elif label.lower() == "negative": # Your model might output 'LABEL_0' or 'NEG'
        emoji = "😠"
        description = "This text expresses a **highly negative** sentiment." if score > 0.9 else "This text expresses a **negative** sentiment."
        color_class = "sentiment-negative"
    elif label.lower() == "neutral": # Your model might output 'LABEL_1' or 'NEUTRAL'
        emoji = "😐"
        description = "This text expresses a **neutral** sentiment."
        color_class = "sentiment-neutral"
    else:
        emoji = "❓"
        description = "Could not confidently determine sentiment."
        color_class = ""
    
    return f"<div class='sentiment-display {color_class}'>{emoji} {label.upper()} ({score:.2f})</div>" + \
           f"<p>{description}</p>"

# --- Sentiment Analysis Function ---
def analyze_sentiment(text):
    if not model_loaded_successfully:
        return {
            "Overall Sentiment": "<div class='sentiment-display'>⚠️ Model Not Loaded ⚠️</div><p>Please contact the administrator. The sentiment analysis model failed to load.</p>",
            "Confidence Scores": {},
            "Raw Output": "Model loading failed."
        }

    if not text.strip():
        return {
            "Overall Sentiment": "<div class='sentiment-display'>✍️ Please enter some text! ✍️</div><p>Start typing to analyze its sentiment.</p>",
            "Confidence Scores": {},
            "Raw Output": ""
        }

    try:
        # The pipeline returns a list of dictionaries if return_all_scores=True
        # e.g., [[{'label': 'LABEL_0', 'score': 0.9}, {'label': 'LABEL_1', 'score': 0.05}]]
        results = sentiment_analyzer(text)[0] # Get the first (and only) list of results

        # Sort results by score in descending order
        results_sorted = sorted(results, key=lambda x: x['score'], reverse=True)

        # Get the top sentiment
        top_sentiment = results_sorted[0]
        label = top_sentiment['label']
        score = top_sentiment['score']

        # Format for Gradio Label component (Dictionary {label: score})
        # This is for the 'Confidence Scores' output
        confidence_scores_output = {item['label']: item['score'] for item in results}

        # Interpret sentiment for the 'Overall Sentiment' output
        overall_sentiment_display = interpret_sentiment(label, score)

        return {
            "Overall Sentiment": overall_sentiment_display,
            "Confidence Scores": confidence_scores_output,
            "Raw Output": str(results)
        }
    except Exception as e:
        return {
            "Overall Sentiment": f"<div class='sentiment-display'>❌ Error ❌</div><p>An error occurred during analysis: {e}</p>",
            "Confidence Scores": {},
            "Raw Output": f"Error: {e}"
        }

# --- Gradio Interface ---
with gr.Blocks(css=custom_css, theme=gr.themes.Soft()) as demo: # Using gr.Blocks for more layout control
    gr.Markdown("# ✨ Sentiment Spark ✨")
    gr.Markdown("---")
    gr.Markdown("### Uncover the emotional tone of your English text instantly!")

    with gr.Row(): # Horizontal layout for input and outputs
        with gr.Column(scale=2):
            text_input = gr.Textbox(
                lines=7,
                placeholder="Type your English text here...",
                label="Your Text",
                interactive=True,
                value="This movie was absolutely brilliant! A masterpiece of storytelling and emotion."
            )
            analyze_btn = gr.Button("Analyze Sentiment", variant="primary") # The main action button

            gr.Markdown("---")
            gr.Markdown("### Try some examples:")
            gr.Examples(
                examples=[
                    ["This product exceeded my expectations, truly amazing!"],
                    ["I found the customer service to be quite disappointing and slow."],
                    ["The weather forecast predicts light rain for tomorrow morning."],
                    ["What a fantastic experience! Highly recommend it."],
                    ["I'm so frustrated with this slow internet connection."],
                    ["The meeting concluded without any major decisions."]
                ],
                inputs=text_input,
                cache_examples=True # Caching examples for faster loading
            )

        with gr.Column(scale=3):
            gr.Markdown("## 📈 Analysis Results 📉")
            overall_sentiment_output = gr.HTML(label="Overall Sentiment") # Using HTML to render custom styled sentiment
            confidence_scores_output = gr.Label(num_top_classes=3, label="Confidence Scores") # Default Gradio Label
            raw_output = gr.JSON(label="Raw Model Output", visible=False) # For debugging/advanced users

    # --- Event Listeners ---
    text_input.change(
        fn=analyze_sentiment,
        inputs=text_input,
        outputs=[overall_sentiment_output, confidence_scores_output, raw_output],
        # live=True # Uncomment for live updates as user types (can be resource intensive)
    )
    analyze_btn.click(
        fn=analyze_sentiment,
        inputs=text_input,
        outputs=[overall_sentiment_output, confidence_scores_output, raw_output]
    )

demo.launch()