andrej-1985/HyperparameterTunning

1"""2🚀 SIMPLE ML FRONTEND - DIRECT INTEGRATION3=========================================4Benutzer setzt Parameter → Training startet → Ergebnisse anzeigen5"""6 7import streamlit as st8import plotly.graph_objects as go9import plotly.express as px10import pandas as pd11import numpy as np12import time13import json14from typing import Dict, Any15import threading16import queue17 18# Import deiner Original-Klassen (angepasst)19import os20import logging21from sklearn.datasets import load_diabetes22from sklearn.model_selection import train_test_split23from sklearn.preprocessing import StandardScaler24from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score as sklearn_r225import tensorflow as tf26import keras27from keras.optimizers import Adam28from keras.layers import Dense, Dropout29from keras.models import Sequential30from keras.callbacks import EarlyStopping, ReduceLROnPlateau, ModelCheckpoint31from keras.regularizers import l232from datetime import datetime33 34# Setup Logging35logging.basicConfig(level=logging.INFO)36logger = logging.getLogger(__name__)37 38class StreamlitMLTrainer:39    """40    🎯 VEREINFACHTE ML-KLASSE FÜR STREAMLIT INTEGRATION41    Basiert auf deiner Original-Logik + UI-Updates42    """43 44    def __init__(self):45        self.model = None46        self.scaler = None47        self.y_scaler = None48        self.history = None49        self.training_active = False50 51        # UI Update Queue52        if 'training_queue' not in st.session_state:53            st.session_state.training_queue = queue.Queue()54        if 'training_logs' not in st.session_state:55            st.session_state.training_logs = []56 57    def log_to_ui(self, message: str):58        """📝 Sende Log-Message an UI"""59        st.session_state.training_logs.append(f"{datetime.now().strftime('%H:%M:%S')} - {message}")60        # Nur letzte 50 Logs behalten61        if len(st.session_state.training_logs) > 50:62            st.session_state.training_logs = st.session_state.training_logs[-50:]63 64    def load_and_preprocess_data(self, test_size: float = 0.2, val_size: float = 0.2):65        """📊 Daten laden und vorverarbeiten"""66        self.log_to_ui("📊 Lade Diabetes-Dataset...")67 68        # Original Logik aus deinem Code69        dataset = load_diabetes()70        X, y = dataset.data, dataset.target.reshape(-1, 1)71 72        self.log_to_ui(f"✅ Dataset geladen: {X.shape[0]} Samples, {X.shape[1]} Features")73 74        # Splits75        X_temp, X_test, y_temp, y_test = train_test_split(76            X, y, test_size=test_size, random_state=4277        )78        X_train, X_val, y_train, y_val = train_test_split(79            X_temp, y_temp, test_size=val_size, random_state=4280        )81 82        # Standardisierung83        self.scaler = StandardScaler()84        X_train_scaled = self.scaler.fit_transform(X_train).astype(np.float32)85        X_val_scaled = self.scaler.transform(X_val).astype(np.float32)86        X_test_scaled = self.scaler.transform(X_test).astype(np.float32)87 88        self.y_scaler = StandardScaler()89        y_train_scaled = self.y_scaler.fit_transform(y_train).astype(np.float32)90        y_val_scaled = self.y_scaler.transform(y_val).astype(np.float32)91        y_test_scaled = self.y_scaler.transform(y_test).astype(np.float32)92 93        self.log_to_ui(f"🔄 Daten verarbeitet: {len(X_train)} Train, {len(X_val)} Val, {len(X_test)} Test")94 95        return (X_train_scaled, y_train_scaled), (X_val_scaled, y_val_scaled), (X_test_scaled, y_test_scaled)96 97    def build_model(self, hidden_layers: list, dropout_rate: float, l2_reg: float, learning_rate: float):98        """🏗️ Modell erstellen"""99        self.log_to_ui(f"🏗️ Erstelle Modell: {len(hidden_layers)} Hidden Layers {hidden_layers}")100 101        model = Sequential(name="StreamlitDiabetesRegression")102 103        # Hidden Layers104        for i, units in enumerate(hidden_layers):105            if i == 0:106                model.add(Dense(107                    units=units, input_shape=(10,),108                    activation='relu', kernel_initializer='he_normal',109                    kernel_regularizer=l2(l2_reg)110                ))111            else:112                model.add(Dense(113                    units=units, activation='relu',114                    kernel_initializer='he_normal',115                    kernel_regularizer=l2(l2_reg)116                ))117 118            model.add(Dropout(dropout_rate))119 120        # Output Layer121        model.add(Dense(1, kernel_initializer='he_normal'))122 123        # Kompilieren124        model.compile(125            optimizer=Adam(learning_rate=learning_rate),126            loss='mse',127            metrics=['mae']128        )129 130        total_params = model.count_params()131        self.log_to_ui(f"✅ Modell erstellt: {total_params:,} Parameter")132 133        return model134 135    def train_model_with_ui_updates(self, model, train_data, val_data,136                                   max_epochs: int, batch_size: int, patience: int):137        """🚀 Training mit Live-UI-Updates"""138 139        X_train, y_train = train_data140        X_val, y_val = val_data141 142        self.log_to_ui(f"🚀 Starte Training: {max_epochs} max Epochs, Batch Size {batch_size}")143 144        # Custom Callback für UI-Updates145        class StreamlitCallback(keras.callbacks.Callback):146            def __init__(self, ui_logger):147                self.ui_logger = ui_logger148                self.start_time = time.time()149 150            def on_epoch_end(self, epoch, logs=None):151                # Nur jede 5. Epoch für Performance152                if (epoch + 1) % 5 == 0 or epoch < 10:153                    elapsed = time.time() - self.start_time154                    self.ui_logger(155                        f"📈 Epoch {epoch+1}: Loss={logs['loss']:.4f}, "156                        f"Val_Loss={logs['val_loss']:.4f}, Zeit={elapsed:.1f}s"157                    )158 159            def on_train_end(self, logs=None):160                total_time = time.time() - self.start_time161                self.ui_logger(f"✅ Training beendet nach {total_time:.1f} Sekunden")162 163        # Callbacks164        callbacks = [165            StreamlitCallback(self.log_to_ui),166            EarlyStopping(monitor='val_loss', patience=patience, restore_best_weights=True, verbose=0),167            ReduceLROnPlateau(monitor='val_loss', factor=0.5, patience=patience//2, min_lr=1e-7, verbose=0)168        ]169 170        # Training171        self.history = model.fit(172            X_train, y_train,173            batch_size=batch_size,174            epochs=max_epochs,175            validation_data=(X_val, y_val),176            callbacks=callbacks,177            verbose=0  # UI übernimmt Output178        )179 180        return model181 182    def evaluate_model_with_ui(self, model, test_data):183        """📊 Evaluation mit UI-Feedback"""184        self.log_to_ui("🎯 Starte Modell-Evaluierung...")185 186        X_test, y_test = test_data187        y_pred_scaled = model.predict(X_test, verbose=0)188 189        # Zurück-transformieren für echte Metriken190        y_test_original = self.y_scaler.inverse_transform(y_test)191        y_pred_original = self.y_scaler.inverse_transform(y_pred_scaled)192 193        # Metriken berechnen194        results = {195            'mse': float(mean_squared_error(y_test_original, y_pred_original)),196            'rmse': float(np.sqrt(mean_squared_error(y_test_original, y_pred_original))),197            'mae': float(mean_absolute_error(y_test_original, y_pred_original)),198            'r2': float(sklearn_r2(y_test_original, y_pred_original)),199            'mape': float(np.mean(np.abs((y_test_original - y_pred_original) / y_test_original)) * 100),200            'test_samples': len(y_test),201            'y_true': y_test_original.flatten(),202            'y_pred': y_pred_original.flatten()203        }204 205        self.log_to_ui(f"🎯 Evaluierung abgeschlossen: R²={results['r2']:.4f}, MSE={results['mse']:.1f}")206 207        return results208 209    def run_complete_training(self, config: Dict[str, Any]) -> Dict[str, Any]:210        """🎯 Komplettes Training Pipeline"""211        try:212            # Reset213            st.session_state.training_logs = []214            self.training_active = True215 216            # 1. Daten laden217            train_data, val_data, test_data = self.load_and_preprocess_data(218                config['test_size'], config['validation_size']219            )220 221            # 2. Modell erstellen222            self.model = self.build_model(223                config['hidden_layers'],224                config['dropout_rate'],225                config['l2_reg'],226                config['learning_rate']227            )228 229            # 3. Training230            self.model = self.train_model_with_ui_updates(231                self.model, train_data, val_data,232                config['max_epochs'], config['batch_size'], config['patience']233            )234 235            # 4. Evaluation236            results = self.evaluate_model_with_ui(self.model, test_data)237 238            self.training_active = False239            self.log_to_ui("🏆 Pipeline komplett abgeschlossen!")240 241            return results242 243        except Exception as e:244            self.training_active = False245            self.log_to_ui(f"❌ Fehler: {e}")246            return {"error": str(e)}247 248# ============================================================================249# STREAMLIT UI APPLICATION250# ============================================================================251 252def main():253    """🚀 Hauptanwendung"""254 255    st.set_page_config(256        page_title="🚀 ML Hyperparameter Tuning",257        page_icon="🚀",258        layout="wide"259    )260 261    st.title("🚀 ML Hyperparameter Tuning & Training")262    st.markdown("**Set Parameters → Start Training → View Results**")263 264    # Initialize Trainer265    if 'trainer' not in st.session_state:266        st.session_state.trainer = StreamlitMLTrainer()267 268    # ⚙️ SIDEBAR: Hyperparameter Configuration269    st.sidebar.header("⚙️ Hyperparameter Configuration")270 271    # Model Architecture272    with st.sidebar.expander("🏗️ Model Architecture", expanded=True):273        num_layers = st.slider("Anzahl Hidden Layers", 1, 5, 3)274 275        hidden_layers = []276        for i in range(num_layers):277            units = st.slider(278                f"Layer {i+1} Neurons", 8, 256,279                [128, 64, 32, 16, 8][i] if i < 5 else 16,280                step=8, key=f"layer_{i}"281            )282            hidden_layers.append(units)283 284    # Regularization285    with st.sidebar.expander("🛡️ Regularization", expanded=True):286        dropout_rate = st.slider("Dropout Rate", 0.0, 0.8, 0.3, 0.05)287        l2_reg = st.slider("L2 Regularization", 0.0, 0.1, 0.01, 0.005)288 289    # Training Parameters290    with st.sidebar.expander("🎯 Training Parameters", expanded=True):291        learning_rate = st.selectbox("Learning Rate", [0.01, 0.005, 0.001, 0.0005], index=2)292        batch_size = st.selectbox("Batch Size", [16, 32, 64, 128], index=1)293        max_epochs = st.slider("Max Epochs", 100, 2000, 1000, 100)294        patience = st.slider("Early Stopping Patience", 10, 100, 50, 10)295        test_size = st.slider("Test Size", 0.1, 0.3, 0.2, 0.05)296        validation_size = st.slider("Validation Size", 0.1, 0.3, 0.2, 0.05)297 298    # Sammle alle Parameter299    config = {300        'hidden_layers': hidden_layers,301        'dropout_rate': dropout_rate,302        'l2_reg': l2_reg,303        'learning_rate': learning_rate,304        'batch_size': batch_size,305        'max_epochs': max_epochs,306        'patience': patience,307        'test_size': test_size,308        'validation_size': validation_size309    }310 311    # 🏗️ MAIN AREA: Tabs für verschiedene Bereiche312    tab1, tab2, tab3 = st.tabs(["🏗️ Model Preview", "🚀 Training", "📊 Results"])313 314    # ============================================================================315    # TAB 1: MODEL PREVIEW316    # ============================================================================317    with tab1:318        st.header("🏗️ Model Architecture Preview")319 320        col1, col2 = st.columns([2, 1])321 322        with col1:323            # Visualisiere Architektur324            fig = go.Figure()325 326            layers = ["Input (10)"] + [f"Hidden {i+1} ({units})" for i, units in enumerate(hidden_layers)] + ["Output (1)"]327 328            for i, layer_name in enumerate(layers):329                color = "lightblue" if i == 0 else ("lightcoral" if i == len(layers)-1 else "lightgreen")330 331                fig.add_shape(332                    type="rect", x0=0, y0=i*1.2, x1=3, y1=i*1.2+1,333                    fillcolor=color, line=dict(color="black", width=1)334                )335 336                fig.add_annotation(337                    x=1.5, y=i*1.2+0.5, text=layer_name,338                    showarrow=False, font=dict(size=12)339                )340 341            fig.update_layout(342                title="🏗️ Neural Network Architecture",343                xaxis=dict(visible=False), yaxis=dict(visible=False),344                height=400, showlegend=False345            )346 347            st.plotly_chart(fig, use_container_width=True)348 349        with col2:350            # Parameter Summary351            total_params = sum([10 * hidden_layers[0]] +352                             [hidden_layers[i] * hidden_layers[i+1] for i in range(len(hidden_layers)-1)] +353                             [hidden_layers[-1]])354 355            st.metric("🔢 Est. Parameters", f"{total_params:,}")356            st.metric("🛡️ Dropout Rate", f"{dropout_rate:.1%}")357            st.metric("⚡ Learning Rate", f"{learning_rate}")358            st.metric("📦 Batch Size", batch_size)359            st.metric("🎯 Max Epochs", max_epochs)360            st.metric("⏸️ Patience", patience)361 362    # ============================================================================363    # TAB 2: TRAINING INTERFACE364    # ============================================================================365    with tab2:366        st.header("🚀 Training Interface")367 368        # Training Button369        col1, col2, col3 = st.columns([1, 1, 1])370 371        with col1:372            if st.button("🚀 Start Training", type="primary",373                        disabled=st.session_state.get('training_active', False)):374 375                st.session_state.training_active = True376                st.session_state.training_logs = []377                st.session_state.results = None378 379                # Starte Training380                with st.spinner("🔄 Training läuft..."):381                    results = st.session_state.trainer.run_complete_training(config)382                    st.session_state.results = results383                    st.session_state.training_active = False384 385                if 'error' not in results:386                    st.success("✅ Training erfolgreich abgeschlossen!")387                    st.balloons()  # 🎉 Celebration!388                else:389                    st.error(f"❌ Training fehlgeschlagen: {results['error']}")390 391        with col2:392            if st.button("🧹 Clear Logs"):393                st.session_state.training_logs = []394 395        with col3:396            if st.button("⏹️ Stop Training"):397                st.session_state.training_active = False398                st.warning("⏸️ Training gestoppt (nicht implementiert)")399 400        # Training Status401        if st.session_state.get('training_active', False):402            st.info("🔄 Training läuft... Bitte warten.")403 404            # Progress Animation405            progress_bar = st.progress(0)406            status_text = st.empty()407 408            # Simuliere Progress (da echtes Training zu schnell für UI)409            for i in range(100):410                progress_bar.progress(i + 1)411                status_text.text(f"Training Progress: {i+1}%")412                time.sleep(0.05)  # 5 Sekunden total413 414        # Live Training Logs415        st.subheader("📋 Training Logs")416 417        if st.session_state.training_logs:418            # Container für Logs (Auto-Scroll)419            log_container = st.container()420            with log_container:421                # Zeige neueste Logs zuerst422                for log in reversed(st.session_state.training_logs[-20:]):423                    st.text(log)424        else:425            st.info("🎯 Klicke 'Start Training' um Logs zu sehen...")426 427    # ============================================================================428    # TAB 3: RESULTS DASHBOARD429    # ============================================================================430    with tab3:431        st.header("📊 Training Results Dashboard")432 433        if not st.session_state.get('results') or 'error' in st.session_state.get('results', {}):434            st.info("🎯 Starte Training um Ergebnisse zu sehen...")435            return436 437        results = st.session_state.results438 439        # 🏆 Key Metrics440        st.subheader("🏆 Performance Metrics")441 442        col1, col2, col3, col4, col5 = st.columns(5)443 444        with col1:445            delta_r2 = f"+{(results['r2'] - 0.49)*100:.1f}%" if results['r2'] > 0.49 else None446            st.metric("🎯 R² Score", f"{results['r2']:.4f}", delta_r2)447 448        with col2:449            st.metric("📉 MSE", f"{results['mse']:.1f}")450 451        with col3:452            st.metric("📏 RMSE", f"{results['rmse']:.1f}")453 454        with col4:455            st.metric("📊 MAE", f"{results['mae']:.1f}")456 457        with col5:458            st.metric("📈 MAPE", f"{results['mape']:.1f}%")459 460        # 📈 Visualizations461        col1, col2 = st.columns(2)462 463        with col1:464            # Prediction vs Actual Scatter Plot465            fig_scatter = px.scatter(466                x=results['y_true'], y=results['y_pred'],467                labels={'x': 'Actual Values', 'y': 'Predicted Values'},468                title="🎯 Predictions vs Actual Values"469            )470 471            # Perfekte Vorhersage-Linie472            min_val, max_val = min(results['y_true']), max(results['y_true'])473            fig_scatter.add_trace(go.Scatter(474                x=[min_val, max_val], y=[min_val, max_val],475                mode='lines', name='Perfect Prediction',476                line=dict(color='red', dash='dash')477            ))478 479            st.plotly_chart(fig_scatter, use_container_width=True)480 481        with col2:482            # Performance Gauge483            fig_gauge = go.Figure(go.Indicator(484                mode="gauge+number+delta",485                value=results['r2'],486                domain={'x': [0, 1], 'y': [0, 1]},487                title={'text': "🎯 R² Performance"},488                delta={'reference': 0.49},489                gauge={490                    'axis': {'range': [0, 1]},491                    'bar': {'color': "darkblue"},492                    'steps': [493                        {'range': [0, 0.4], 'color': "lightgray"},494                        {'range': [0.4, 0.6], 'color': "yellow"},495                        {'range': [0.6, 0.8], 'color': "lightgreen"},496                        {'range': [0.8, 1.0], 'color': "green"}497                    ],498                    'threshold': {499                        'line': {'color': "red", 'width': 4},500                        'thickness': 0.75, 'value': 0.65501                    }502                }503            ))504 505            st.plotly_chart(fig_gauge, use_container_width=True)506 507        # 📊 Training History508        if st.session_state.trainer.history:509            st.subheader("📈 Training History")510 511            history_df = pd.DataFrame(st.session_state.trainer.history.history)512 513            fig_history = go.Figure()514            fig_history.add_trace(go.Scatter(515                x=list(range(len(history_df))),516                y=history_df['loss'],517                mode='lines',518                name='Training Loss',519                line=dict(color='blue')520            ))521            fig_history.add_trace(go.Scatter(522                x=list(range(len(history_df))),523                y=history_df['val_loss'],524                mode='lines',525                name='Validation Loss',526                line=dict(color='red')527            ))528 529            fig_history.update_layout(530                title="📈 Training & Validation Loss",531                xaxis_title="Epochs",532                yaxis_title="Loss",533                height=400534            )535 536            st.plotly_chart(fig_history, use_container_width=True)537 538        # 📋 Detailed Results Table539        st.subheader("📋 Detailed Results")540 541        results_data = {542            'Metric': ['R² Score', 'Mean Squared Error', 'Root MSE', 'Mean Absolute Error', 'MAPE', 'Test Samples'],543            'Value': [544                f"{results['r2']:.4f}",545                f"{results['mse']:.2f}",546                f"{results['rmse']:.2f}",547                f"{results['mae']:.2f}",548                f"{results['mape']:.1f}%",549                f"{results['test_samples']}"550            ],551            'Status': [552                '🎯 Excellent' if results['r2'] > 0.6 else '✅ Good' if results['r2'] > 0.4 else '⚠️ Poor',553                '✅ Good' if results['mse'] < 3000 else '⚠️ High',554                '✅ Good' if results['rmse'] < 55 else '⚠️ High',555                '✅ Good' if results['mae'] < 45 else '⚠️ High',556                '✅ Good' if results['mape'] < 20 else '⚠️ High',557                '📊 Info'558            ]559        }560 561        st.dataframe(pd.DataFrame(results_data), use_container_width=True)562 563        # 💾 Download Results564        if st.button("💾 Download Results JSON"):565            results_json = json.dumps({566                'config': config,567                'results': results,568                'timestamp': datetime.now().isoformat()569            }, indent=2)570 571            st.download_button(572                "📁 Download",573                results_json,574                f"ml_results_{int(time.time())}.json",575                "application/json"576            )577 578if __name__ == "__main__":579    main()580 581"""582🎯 SIMPLE & DIRECT APPROACH:583===========================584 585✅ **Eine Datei** - Alles in einem586✅ **Direkte Integration** - Keine komplexe MVC-Struktur587✅ **Live Training Updates** - Callback direkt in UI588✅ **Automatisches Training** - Button → Training → Ergebnisse589✅ **Schöne Results UI** - Plots, Metriken, Downloads590✅ **Deine Original-Logik** - Komplett erhalten591 592🚀 **Usage:**5931. Speichern als: streamlit_ml_app.py5942. Run: streamlit run streamlit_ml_app.py5953. Set Parameters → Click Training → View Results596 597💡 **Features:**598- Dynamic Hyperparameter Tuning599- Real-time Training Logs600- Interactive Results Dashboard601- Model Architecture Visualization602- Results Export Functionality603- Progress Tracking604"""605