import os, json, joblib, numpy as np, pandas as pd from datetime import datetime from flask import current_app from sklearn.model_selection import train_test_split from sklearn.preprocessing import LabelEncoder, StandardScaler, OrdinalEncoder from sklearn.compose import ColumnTransformer from sklearn.pipeline import Pipeline from sklearn.ensemble import RandomForestClassifier from sklearn.multioutput import ClassifierChain from sklearn.metrics import accuracy_score, classification_report from app.extensions import db from app.models import MLModel, TrainingHistory FITUR_STRESS = [f"S{i}" for i in range(1, 41)] FITUR_MOTIVASI = [f"M{i}" for i in range(1, 29)] FITUR_DEMO = ['Jurusan', 'Angkatan', 'Gender', 'Usia', 'IPK', 'freq_olahraga', 'durasi_tidur'] def train_model(history_id, csv_path, model_type, hyperparams=None, models_folder="models"): history = TrainingHistory.query.get(history_id) if not history: return try: history.status = "running" history.progress_message = json.dumps({"progress": 10, "message": "Membaca dataset..."}) db.session.commit() df = pd.read_csv(csv_path) if csv_path.endswith('.csv') else pd.read_excel(csv_path) # ── 1. Label stress ────────────────────────────────────── stress_cols = [f"S{i}" for i in range(1, 41)] df['total_stress'] = df[stress_cols].sum(axis=1) df['label_stress'], bins_stress = pd.qcut( df['total_stress'], q=3, labels=["Rendah", "Sedang", "Tinggi"], retbins=True ) # ── 2. Label motivasi (SDI) ────────────────────────────── df['mean_im_know'] = df[['M1','M2','M3','M4']].mean(axis=1) df['mean_im_acc'] = df[['M5','M6','M7','M8']].mean(axis=1) df['mean_im_stim'] = df[['M9','M10','M11','M12']].mean(axis=1) df['mean_identified'] = df[['M13','M14','M15','M16']].mean(axis=1) df['mean_introjected'] = df[['M17','M18','M19','M20']].mean(axis=1) df['mean_external'] = df[['M21','M22','M23','M24']].mean(axis=1) df['mean_amotivation'] = df[['M25','M26','M27','M28']].mean(axis=1) df['intrinsic_total'] = (df['mean_im_know'] + df['mean_im_acc'] + df['mean_im_stim']) / 3 df['controlled_extrinsic'] = (df['mean_introjected'] + df['mean_external']) / 2 df['score_sdi'] = (2 * df['intrinsic_total']) + (1 * df['mean_identified']) \ - (1 * df['controlled_extrinsic']) - (2 * df['mean_amotivation']) df['label_motivasi'], bins_motivasi = pd.qcut( df['score_sdi'], q=3, labels=["Rendah", "Sedang", "Tinggi"], retbins=True ) # ── 3. Encode label ────────────────────────────────────── le_stress = LabelEncoder() le_motivasi = LabelEncoder() y_stress = le_stress.fit_transform(df['label_stress']) y_motivasi = le_motivasi.fit_transform(df['label_motivasi']) y = np.column_stack([y_stress, y_motivasi]) # ── 4. Features ────────────────────────────────────────── X = df[FITUR_STRESS + FITUR_MOTIVASI + FITUR_DEMO].copy() # ── 5. Encode kategorik demo ───────────────────────────── le_jurusan = LabelEncoder() le_gender = LabelEncoder() le_olahraga = LabelEncoder() le_tidur = LabelEncoder() df['Jurusan'] = le_jurusan.fit_transform(df['Jurusan'].astype(str)) df['Gender'] = le_gender.fit_transform(df['Gender'].astype(str)) df['freq_olahraga'] = le_olahraga.fit_transform(df['freq_olahraga'].astype(str)) df['durasi_tidur'] = le_tidur.fit_transform(df['durasi_tidur'].astype(str)) cats_jurusan = list(le_jurusan.classes_) cats_gender = list(le_gender.classes_) cats_olahraga = list(le_olahraga.classes_) cats_tidur = list(le_tidur.classes_) demo_encoder = OrdinalEncoder( categories=[cats_jurusan, cats_gender, cats_olahraga, cats_tidur], handle_unknown='use_encoded_value', unknown_value=-1 ) preprocessor = ColumnTransformer([ ('scaler_stress', StandardScaler(), FITUR_STRESS), ('scaler_motivasi', StandardScaler(), FITUR_MOTIVASI), ('demo_encoder', demo_encoder, ['Jurusan', 'Gender', 'freq_olahraga', 'durasi_tidur']) ], remainder='passthrough') # ── 6. Split ───────────────────────────────────────────── X_train, X_test, y_train, y_test = train_test_split( X, y, test_size=0.2, random_state=42, stratify=y[:, 0] ) # ── 7. Hyperparams ─────────────────────────────────────── params = { 'n_estimators': 100, 'max_depth': 10, 'min_samples_leaf': 3, 'max_features': 'sqrt', 'class_weight': 'balanced', 'random_state': 42 } if hyperparams and 'params' in hyperparams: params.update(hyperparams['params']) base_rf = RandomForestClassifier(**params) # ClassifierChain dengan order [1,0] (motivasi dulu, lalu stress) chain_rf = ClassifierChain(base_rf, order=[1, 0], random_state=42) pipeline_chain = Pipeline([ ('preprocess', preprocessor), ('clf', chain_rf) ]) history.progress_message = json.dumps({"progress": 50, "message": "Melatih model..."}) db.session.commit() pipeline_chain.fit(X_train, y_train) y_pred = pipeline_chain.predict(X_test) # ── 8. Evaluasi per label ──────────────────────────────── acc_stress = accuracy_score(y_test[:, 0], y_pred[:, 0]) acc_motivasi = accuracy_score(y_test[:, 1], y_pred[:, 1]) exact_match = np.mean(np.all(y_test == y_pred, axis=1)) report_stress = classification_report(y_test[:, 0], y_pred[:, 0], target_names=['Rendah','Sedang','Tinggi'], output_dict=True) report_motivasi = classification_report(y_test[:, 1], y_pred[:, 1], target_names=['Rendah','Sedang','Tinggi'], output_dict=True) metrics = { 'exact_match': round(exact_match, 4), 'accuracy_stress': round(acc_stress, 4), 'accuracy_motivasi': round(acc_motivasi, 4), 'report_stress': report_stress, 'report_motivasi': report_motivasi } # ── 9. Simpan artifact ─────────────────────────────────── os.makedirs(models_folder, exist_ok=True) timestamp = datetime.utcnow().strftime('%Y%m%d%H%M%S') uid = history.task_id[:8] model_path = os.path.join(models_folder, f"model_chain_{timestamp}_{uid}.joblib") # Konversi bins ke list bins_stress_list = bins_stress.tolist() bins_motivasi_list = bins_motivasi.tolist() artifact = { 'pipeline': pipeline_chain, 'feature_names': FITUR_STRESS + FITUR_MOTIVASI + FITUR_DEMO, 'bins_stress': bins_stress_list, 'bins_motivasi': bins_motivasi_list, 'label_map': {0: 'Rendah', 1: 'Sedang', 2: 'Tinggi'}, 'demo_categories': { 'Jurusan': cats_jurusan, 'Gender': cats_gender, 'freq_olahraga': cats_olahraga, 'durasi_tidur': cats_tidur }, 'params': params, 'chain_order': [1, 0] # urutan motivasi dulu lalu stres } joblib.dump(artifact, model_path) # ── 10. Simpan dua record model ke database ────────────── for mtype in ['stress', 'motivasi']: rep = report_stress if mtype == 'stress' else report_motivasi precision = round(rep['weighted avg']['precision'], 4) recall = round(rep['weighted avg']['recall'], 4) f1 = round(rep['weighted avg']['f1-score'], 4) model_record = MLModel( type=mtype, algorithm='ClassifierChain_RF', version=f"v_{timestamp}", file_path=model_path, is_active=False, accuracy=metrics[f'accuracy_{mtype}'], precision_score=precision, recall_score=recall, f1_score=f1, data_count=len(df), qcut_thresholds=json.dumps({ 'bins_stress': bins_stress_list, 'bins_motivasi': bins_motivasi_list }), artifact_metadata=json.dumps({ 'bins_stress': bins_stress_list, 'bins_motivasi': bins_motivasi_list, 'demo_categories': artifact['demo_categories'], 'best_params': params, 'cv_macro_f1': None, 'chain_order': [1, 0] }), created_at=datetime.utcnow() ) db.session.add(model_record) db.session.flush() if mtype == model_type: history.model_id = model_record.id history.status = "completed" history.metrics = json.dumps(metrics) history.progress_message = json.dumps({"progress": 100, "message": "Selesai"}) db.session.commit() except Exception as e: current_app.logger.error(f"Training failed: {e}") history.status = "failed" history.progress_message = json.dumps({"progress": 0, "message": str(e)}) db.session.commit()