import gradio as gr
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
import json
import os

# --- 1. CONFIGURATION ---
MODEL_PATH = "rup69/Sentiment-Analysis" 
LABELS = ["anger", "fear", "joy", "sadness", "surprise"]
DEFAULT_THRESHOLDS = {"anger": 0.5, "fear": 0.5, "joy": 0.5, "sadness": 0.5, "surprise": 0.5}

# --- 2. LOAD RESOURCES ---
# Load Model
try:
    model = AutoModelForSequenceClassification.from_pretrained(MODEL_PATH)
    print("✅ Model loaded from Model Repo!")
except Exception as e:
    model = None
    print(f"❌ Error loading model: {e}")

# Load Tokenizer
try:
    tokenizer = AutoTokenizer.from_pretrained("roberta-large")
except:
    tokenizer = AutoTokenizer.from_pretrained("roberta-base")

# Load Thresholds
if os.path.exists("fine_tune_thresholds.json"):
    with open("fine_tune_thresholds.json", "r") as f:
        THRESHOLDS = json.load(f)
else:
    THRESHOLDS = DEFAULT_THRESHOLDS

# --- 3. LOGIC ---
def analyze(text):
    # 1. Simple Check: Is the model variable loaded?
    if model is None: 
        return {}, "⚠️ Model failed to load. Check logs."

    # 2. Check for Empty Text
    if not text or not text.strip():
        return {}, "⚠️ Please type something first!"

    inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)
    model.eval()
    with torch.no_grad():
        outputs = model(**inputs)
        probs = torch.sigmoid(outputs.logits).numpy()[0]
    
    results = {}
    detected_emotions = []
    
    # Penalties for "Sleep" bias
    penalties = {"sleep": {"fear": 0.50}, "tired": {"fear": 0.40}, "hungry": {"fear": 0.40}}
    text_lower = text.lower()

    for i, label in enumerate(LABELS):
        score = float(probs[i])
        for word, penalty_map in penalties.items():
            if word in text_lower and label in penalty_map:
                score = max(0.0, score - penalty_map[label])
        
        results[label] = score
        # Use simple 0.5 threshold
        if score > 0.5:
            detected_emotions.append(label.upper())
            
    if not detected_emotions:
        final_verdict = "😐 No strong emotion detected."
    else:
        final_verdict = " ".join([f"**{e}**" for e in detected_emotions])
        
    return results, final_verdict

# --- 4. PROFESSIONAL CSS (The "Form" Look) ---
custom_css = """
/* 1. Background Gradient */
.gradio-container {
    background: linear-gradient(135deg, #e0e7ff 0%, #f3f4f6 100%) !important;
}

/* 2. The White Card Container */
.form-card {
    max-width:600px;
    margin:auto;
    background: white !important;
    border-radius: 12px !important;
    padding: 30px !important;
    box-shadow: 0 10px 30px rgba(0,0,0,0.1) !important;
    border-top: 6px solid #6366f1 !important; /* Indigo Top Bar */
    border: 1px solid #e5e7eb !important;
}

/* 3. Header Text */
.form-header h1 {
    color: #4338ca;
    text-align: center;
    font-weight: 800;
    margin-bottom: 5px;
}
.form-header p {
    color: #6b7280;
    text-align: center;
    margin-bottom: 20px;
}

/* 4. Button Styling */
button.primary {
    background: #4f46e5 !important;
    font-size: 1.1em !important;
}
"""

# --- 5. UI LAYOUT ---
theme = gr.themes.Base(primary_hue="indigo", font=[gr.themes.GoogleFont("Inter"), "system-ui"])

with gr.Blocks( title="Sentiment Check") as app:
    
    # We wrap everything in a Column with our custom 'form-card' class
    with gr.Column(elem_classes="form-card"):
        
        # Header
        gr.HTML("""
            <div class="form-header">
                <h1>🧠 Sentiment Check</h1>
                <p>Advanced Neural Emotion Detection System</p>
            </div>
        """)
        
        # Section 1: Input
        gr.Markdown("### 1. Source Text")
        txt_input = gr.Textbox(
            lines=4, 
            placeholder="Type your text here...", 
            show_label=False,
            container=False # Makes it look cleaner
        )
        
        # Action Button
        btn_run = gr.Button("GENERATE REPORT", variant="primary", size="lg")
        
        # Divider
        gr.HTML("<hr style='margin: 20px 0; border: 0; border-top: 1px solid #e5e7eb;'>")
        
        # Section 2: Results
        gr.Markdown("### 2. Analysis Results")
        with gr.Row():
            with gr.Column(scale=1):
                lbl_verdict = gr.Markdown("**Status:** Waiting for data...", label="Prediction")
            with gr.Column(scale=1):
                lbl_chart = gr.Label(num_top_classes=5, label="Confidence Spectrum")
    
    # Footer
    gr.Markdown(
        "<div style='text-align:center; color:#9ca3af; margin-top:15px; font-size:0.8em;'>Secure Local Deployment • Powered by RoBERTa</div>"
    )

    # Logic
    btn_run.click(fn=analyze, inputs=txt_input, outputs=[lbl_chart, lbl_verdict])

app.launch(theme=theme, css=custom_css,share=False)#,auth=("admin", "pass123"))