app.py

import gradio as gr
from google_play_scraper import Sort, reviews
from transformers import pipeline
from collections import Counter
import matplotlib
import threading # <--- Tambahan Penting
import base64
from io import BytesIO
import torch

# ===============================
# 1. SETUP BACKEND & LOCK
# ===============================
# Set backend ke 'Agg' AGAR TIDAK ERROR DI SERVER
matplotlib.use('Agg') 
import matplotlib.pyplot as plt

# Kunci untuk mencegah Matplotlib crash saat diakses banyak user/thread
plot_lock = threading.Lock()

# ===============================
# 2. SETUP MODEL
# ===============================
device = 0 if torch.cuda.is_available() else -1
MODEL_NAME = "Aardiiiiy/indobertweet-base-Indonesian-sentiment-analysis"

print("⏳ Sedang memuat model... Harap tunggu sebentar.")
try:
    sentiment_model = pipeline(
        "text-classification",
        model=MODEL_NAME,
        tokenizer=MODEL_NAME,
        device=device
    )
    print("✅ Model siap digunakan!")
except Exception as e:
    print(f"Error loading model: {e}")

# ===============================
# 3. LOGIC FUNCTIONS
# ===============================
def scrape_reviews(app_id, max_reviews, language):
    try:
        result, _ = reviews(
            app_id,
            lang=language,
            country="id",
            sort=Sort.NEWEST,
            count=max_reviews
        )
        return [r["content"].strip() for r in result if r.get("content")]
    except Exception as e:
        print(f"Error scraping: {e}")
        return []

def analyze_sentiment_batch(texts):
    if not texts:
        return Counter(), 0
    
    try:
        results = sentiment_model(texts, batch_size=16, truncation=True, max_length=512)
        labels = [res["label"].lower() for res in results]
        return Counter(labels), len(labels)
    except Exception as e:
        print(f"Error analyzing: {e}")
        return Counter(), 0

def generate_plot(pos, neu, neg, total):
    # Gunakan LOCK agar thread tidak bertabrakan saat menggambar
    with plot_lock: 
        try:
            colors = ['#4ade80', '#fbbf24', '#f87171']
            labels = ["Positif", "Netral", "Negatif"]
            sizes = [pos, neu, neg]

            fig = plt.figure(figsize=(6, 4))
            fig.patch.set_alpha(0.0) 
            
            plt.pie(
                sizes, 
                labels=labels, 
                autopct="%1.1f%%", 
                startangle=140, 
                colors=colors,
                # Pastikan teks hitam
                textprops={'color':"#000000", 'weight':'bold', 'fontsize': 10},
                wedgeprops={'edgecolor': 'white', 'linewidth': 2}
            )
            
            plt.axis('equal')
            plt.tight_layout()

            buf = BytesIO()
            plt.savefig(buf, format="png", bbox_inches='tight', transparent=True)
            
            # Bersihkan memori segera
            plt.close(fig) 
            plt.close('all') 
            
            return base64.b64encode(buf.getvalue()).decode()
        except Exception as e:
            print(f"Error generating plot: {e}")
            return ""

def generate_html_dashboard(pos, neu, neg, total, img_base64):
    # HTML dengan CSS '!important' agar teks tetap HITAM di tema gelap
    html = f"""
    <div style="
        font-family: 'Segoe UI', sans-serif;
        background-color: #ffffff;
        border-radius: 12px;
        padding: 20px;
        color: #000000 !important; 
        box-shadow: 0 4px 6px rgba(0,0,0,0.1);
    ">
        <h3 style="text-align:center; color:#000000 !important; margin-top:0; font-weight:bold;">
            📊 Hasil Analisis Sentimen
        </h3>
        <p style="text-align:center; color:#333333 !important; font-size:14px;">
            Berdasarkan <b>{total}</b> ulasan terbaru
        </p>

        <div style="display: flex; flex-wrap: wrap; justify-content: center; align-items: center; gap: 20px;">
            <div style="flex: 1; min-width: 250px;">
                <table style="width:100%; color: #000000 !important;">
                    <tr style="border-bottom: 1px solid #eee;">
                        <td style="padding: 8px 0; color: #000000 !important;">🟢 Positif</td>
                        <td style="text-align:right; font-weight:bold; color:#15803d !important;">{pos}</td>
                    </tr>
                    <tr style="border-bottom: 1px solid #eee;">
                        <td style="padding: 8px 0; color: #000000 !important;">🟡 Netral</td>
                        <td style="text-align:right; font-weight:bold; color:#b45309 !important;">{neu}</td>
                    </tr>
                    <tr style="border-bottom: 1px solid #eee;">
                        <td style="padding: 8px 0; color: #000000 !important;">🔴 Negatif</td>
                        <td style="text-align:right; font-weight:bold; color:#b91c1c !important;">{neg}</td>
                    </tr>
                    <tr style="background-color: #f8f9fa;">
                        <td style="padding: 8px 0; font-weight:bold; color: #000000 !important;">Total Data</td>
                        <td style="text-align:right; font-weight:bold; color: #000000 !important;">{total}</td>
                    </tr>
                </table>
            </div>
            <div style="flex: 1; min-width: 250px; text-align: center;">
                <img src="data:image/png;base64,{img_base64}" style="max-width: 100%; height: auto;" />
            </div>
        </div>
    </div>
    """
    return html

def run_pipeline(app_id, max_reviews, language):
    if not app_id:
        return "⚠️ Masukkan Package Name dulu.", "<div></div>"

    texts = scrape_reviews(app_id, max_reviews, language)
    if not texts:
        return "❌ Tidak ditemukan review.", "<div style='color:white'>Data Tidak Ditemukan</div>"

    raw_output = "\n".join(f"[{i+1}] {t}" for i, t in enumerate(texts))
    counter, total = analyze_sentiment_batch(texts)

    pos = counter.get("positive", 0)
    neu = counter.get("neutral", 0)
    neg = counter.get("negative", 0)

    img_base64 = generate_plot(pos, neu, neg, total)
    html_result = generate_html_dashboard(pos, neu, neg, total, img_base64)

    return raw_output, html_result

# ===============================
# 4. GRADIO UI LAYOUT
# ===============================
theme = gr.themes.Soft(
    primary_hue="blue",
    secondary_hue="slate",
).set(
    body_background_fill="#0f172a", 
    block_background_fill="#1e293b", 
    block_label_text_color="#e2e8f0",
    body_text_color="#e2e8f0" 
)

with gr.Blocks(theme=theme, title="Play Store Analyzer") as demo:
    with gr.Row():
        gr.Markdown("# 📱 Google Play Review Analyzer")

    with gr.Group():
        with gr.Row():
            app_id_input = gr.Textbox(label="Package Name", placeholder="com.whatsapp", scale=2)
            language_input = gr.Dropdown(choices=["id", "en"], value="id", label="Bahasa", scale=1)
            max_reviews_input = gr.Slider(minimum=10, maximum=500, step=10, value=50, label="Jumlah", scale=1)
        
        analyze_btn = gr.Button("🔍 Analisis", variant="primary")

    with gr.Row():
        with gr.Column(scale=3):
            analysis_box = gr.HTML(label="Dashboard")
        with gr.Column(scale=2):
            raw_box = gr.Textbox(label="Data Mentah", lines=18)

    analyze_btn.click(
        run_pipeline,
        inputs=[app_id_input, max_reviews_input, language_input],
        outputs=[raw_box, analysis_box]
    )

if __name__ == "__main__":
    # Matikan SSR dan queue untuk stabilitas
    demo.launch(ssr_mode=False)