train_emotion.py

# train_emotion.py
import pandas as pd
import re
import joblib
import os
from datasets import load_dataset
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import LogisticRegression
from sklearn.pipeline import Pipeline
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score, precision_recall_fscore_support

# ==========================================
# 🔧 KONFIGURASI
# ==========================================
MODEL_OUTPUT = 'api/data/model_emotion.pkl' 
# ==========================================

print("🔍 Mengunduh dataset GoEmotions...")

try:
    dataset = load_dataset("google-research-datasets/go_emotions", "simplified", split="train")
    df = pd.DataFrame(dataset)
    labels_list = dataset.features['labels'].feature.names
    
    def get_first_label(label_ids):
        if len(label_ids) > 0:
            return labels_list[label_ids[0]]
        return "neutral"

    df['emotion_label'] = df['labels'].apply(get_first_label)
    X = df['text']
    y = df['emotion_label']
    print(f"✅ Data siap: {len(df)} baris.")

except Exception as e:
    print(f"❌ Error: {e}")
    exit()

# --- CLEANING DATA ---
def clean_text(text):
    text = str(text).lower()
    text = re.sub(r'http\S+', '', text)        
    text = re.sub(r'[^a-zA-Z\s]', '', text)    
    text = re.sub(r'\s+', ' ', text).strip()   
    return text

print("🧹 Membersihkan data emosi...")
X = X.apply(clean_text)

# --- TRAINING ---
print("🚀 Melatih Model Emosi (Logistic Regression Fixed)...")

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

pipeline = Pipeline([
    ('tfidf', TfidfVectorizer(
        max_features=12000,      # Fitur banyak biar detail
        stop_words='english',
        ngram_range=(1, 2),      # Baca kata per kata & frasa
        sublinear_tf=True        # [TRICK] Scaling logaritmik (Penting!)
    )),
    ('clf', LogisticRegression(
        max_iter=1000,           
        solver='lbfgs',          # Ganti ke lbfgs biar aman dari error multiclass
        C=1.2                    # Agak agresif dikit (di atas 1.0) biar akurasi naik
    ))
])

pipeline.fit(X_train, y_train)

# --- EVALUASI ---
print("📊 Menghitung Metrik Evaluasi...")
predictions = pipeline.predict(X_test)

accuracy = accuracy_score(y_test, predictions)
precision, recall, f1, _ = precision_recall_fscore_support(y_test, predictions, average='weighted', zero_division=0)

print("\n" + "="*40)
print("   HASIL EVALUASI MODEL EMOSI (FINAL)")
print("="*40)
print(f"{'Metrik':<15} | {'Skor':<10}")
print("-" * 30)
print(f"{'Akurasi':<15} | {accuracy:.3f} ({accuracy*100:.1f}%)")
print(f"{'Precision':<15} | {precision:.3f}")
print(f"{'Recall':<15} | {recall:.3f}")
print(f"{'F1-Score':<15} | {f1:.3f}")
print("="*40 + "\n")

os.makedirs(os.path.dirname(MODEL_OUTPUT), exist_ok=True)
joblib.dump(pipeline, MODEL_OUTPUT)
print(f"💾 SUKSES! Model Emosi disimpan di: {MODEL_OUTPUT}")