app.py

# ✅ app.py (untuk Gradio Interface di Hugging Face Spaces)

import gradio as gr
import numpy as np
from transformers import AutoTokenizer, TFAutoModelForSequenceClassification
import tensorflow as tf

# Load model dan tokenizer
model_path = "jeanetrixsiee/javo_analisis"
tokenizer = AutoTokenizer.from_pretrained(model_path)
model = TFAutoModelForSequenceClassification.from_pretrained(model_path)

# Mapping label
id2label = model.config.id2label

def predict_sentiment(text):
    try:
        # Tokenisasi input
        inputs = tokenizer(text, return_tensors="tf", padding=True, truncation=True, max_length=256)
        outputs = model(**inputs)
        probs = tf.nn.softmax(outputs.logits, axis=-1).numpy()[0]

        label_id = np.argmax(probs)
        label = id2label[str(label_id)] if str(label_id) in id2label else "Unknown"

        prob_dict = {id2label[str(i)]: float(f"{probs[i]*100:.2f}") for i in range(len(probs))}
        return label, prob_dict

    except Exception as e:
        return "Error", {"Error": str(e)}

# Gradio UI
desc = "Model ini memprediksi sentimen dari komentar YouTube dalam 5 kategori: Very Negative, Negative, Neutral, Positive, Very Positive."

with gr.Blocks() as demo:
    gr.Markdown("## 🔍 Analisis Sentimen Komentar YouTube")
    gr.Markdown(desc)

    with gr.Row():
        with gr.Column():
            input_text = gr.Textbox(label="Masukkan Komentar YouTube")
            clear_btn = gr.Button("Clear")
            submit_btn = gr.Button("Submit", variant="primary")
        with gr.Column():
            output_label = gr.Textbox(label="Hasil Prediksi")
            output_probs = gr.Label(label="Probabilitas Tiap Label (%)")

    submit_btn.click(fn=predict_sentiment, inputs=input_text, outputs=[output_label, output_probs])
    clear_btn.click(lambda: ("", ""), outputs=[output_label, output_probs])

    gr.Markdown("### Share via Link")

# Jalankan
if __name__ == "__main__":
    demo.launch()