app.py

import gradio as gr
import torch
import torch.nn.functional as F
from transformers import RobertaTokenizerFast, RobertaForSequenceClassification
from sklearn.preprocessing import LabelEncoder
import numpy as np

# Load tokenizer dan model
# Load model directly
tokenizer = RobertaTokenizerFast.from_pretrained("FadQ/results")
model = RobertaForSequenceClassification.from_pretrained("FadQ/results")
model.eval()

# Label Ekman (urutan harus cocok dengan urutan training)
ekman_labels = ['anger', 'disgust', 'fear', 'joy', 'sadness', 'surprise', 'neutral']
label_encoder = LabelEncoder()
label_encoder.fit(ekman_labels)

# Fungsi prediksi rata-rata emosi per baris
def predict_emotion_distribution(text):
    lines = [line.strip() for line in text.split('\n') if line.strip()]
    all_probs = []

    for line in lines:
        inputs = tokenizer(line, return_tensors="pt", truncation=True, padding=True, max_length=128)
        with torch.no_grad():
            logits = model(**inputs).logits
            probs = F.softmax(logits, dim=-1).squeeze().cpu().numpy()
            all_probs.append(probs)

    if not all_probs:
        return {label: 0.0 for label in ekman_labels}

    avg_probs = np.mean(all_probs, axis=0)
    result = {label: float(np.round(prob, 4)) for label, prob in zip(ekman_labels, avg_probs)}
    return result

# Gradio Interface
interface = gr.Interface(
    fn=predict_emotion_distribution,
    inputs=gr.Textbox(lines=10, placeholder="Tulis diary-mu. Setiap baris = 1 kalimat...", label="Catatan Harian"),
    outputs=gr.Label(num_top_classes=7, label="Distribusi Emosi"),
    title="Prediksi Emosi dari Diary Harian",
    description="Model mendeteksi emosi (7 label Ekman) dari teks harian. Input dipisah per baris. Output adalah rata-rata probabilitas per emosi."
)

interface.launch()