feat: Красивые графики

src/lab3_pipeline.py

@@ -67,6 +67,11 @@ try:
 except Exception:
     SCIPY_AVAILABLE = False
 
+try:
+    from tbats import TBATS
+    TBATS_AVAILABLE = True
+except ImportError:
+    TBATS_AVAILABLE = False
 
 # -------------------------------------------------------------------------
 # Metrics
@@ -520,6 +525,7 @@ def generate_report_html(out_path: str, plots: List[plt.Figure], tables: Dict[st
 
         # Закрываем рисунок чтобы освободить память
         plt.close(fig)
+        break
 
     html_parts.append("</body></html>")
 
@@ -530,8 +536,26 @@ def generate_report_html(out_path: str, plots: List[plt.Figure], tables: Dict[st
 # -------------------------------------------------------------------------
 # Main runner that orchestrates everything
 # -------------------------------------------------------------------------
+def evaluate_with_cv(models_dict, X, y, cv_method='expanding', n_splits=5):
+    """Оценка моделей с кросс-валидацией"""
+    cv_results = {}
+
+    for name, model_func in models_dict.items():
+        if cv_method == 'expanding':
+            cv_scores = expanding_window_cv(X, y, model_func,
+                                            initial_train_size=len(X) // 2,
+                                            h=30, n_splits=n_splits)
+        else:
+            cv_scores = rolling_window_cv(X, y, model_func,
+                                          window=len(X) // 2,
+                                          h=30, n_splits=n_splits)
+        cv_results[name] = cv_scores
+
+    return cv_results
+
 def run_pipeline(data_path: str, timestamp_col: str, target_col: str,
                  out_report: str = 'lab3_report.html', freq: str = 'D'):
+
     """
     Главная точка запуска pipeline.
     """
@@ -673,6 +697,50 @@ def run_pipeline(data_path: str, timestamp_col: str, target_col: str,
         except Exception as e:
             print("VAR failed:", e)
 
+    # TBATS модель
+    if TBATS_AVAILABLE:
+        try:
+            tbats_model = TBATS(seasonal_periods=[7, 30], use_arma_errors=True)
+            tbats_fitted = tbats_model.fit(y_train)
+            for h in horizons:
+                tbats_pred = tbats_fitted.forecast(steps=h)
+                pred_dates = create_forecast_index(y_train.index[-1], h, inferred_freq)
+                results.append({
+                    'model': 'TBATS',
+                    'h': h,
+                    'pred': pd.Series(tbats_pred, index=pred_dates)
+                })
+        except Exception as e:
+            print("TBATS failed:", e)
+
+    # Prophet модель
+    if PROPHET_AVAILABLE:
+        try:
+            prophet_df = y_train.reset_index()
+            prophet_df.columns = ['ds', 'y']
+            prophet_model = Prophet()
+            prophet_model.fit(prophet_df)
+            future = prophet_model.make_future_dataframe(periods=max(horizons), freq=inferred_freq)
+            forecast = prophet_model.predict(future)
+            for h in horizons:
+                prophet_pred = forecast.tail(h)['yhat'].values
+                pred_dates = create_forecast_index(y_train.index[-1], h, inferred_freq)
+                results.append({
+                    'model': 'Prophet',
+                    'h': h,
+                    'pred': pd.Series(prophet_pred, index=pred_dates)
+                })
+        except Exception as e:
+            print("Prophet failed:", e)
+
+    # GARCH на остатках SARIMAX
+    if ARCH_AVAILABLE and 'sar' in locals():
+        try:
+            garch_model = fit_garch_on_residuals(sar.resid, p=1, q=1)
+            # Прогноз волатильности можно добавить в анализ
+        except Exception as e:
+            print("GARCH failed:", e)
+
     # 3.6 Diagnostics later for top models
     # 3.7 Evaluate on test set
     eval_rows = []
@@ -810,6 +878,11 @@ def run_pipeline(data_path: str, timestamp_col: str, target_col: str,
                 'pred': pd.Series(pred_values, index=pred_dates)
             })
 
+    cv_results = evaluate_with_cv({
+        'SARIMAX': lambda X, y, h: forecast_recursive(fit_sarimax_simple(y), y, h),
+        'AutoARIMA': lambda X, y, h: forecast_recursive(fit_auto_arima(y), y, h)
+    }, df_all.drop(columns=[target_col]), df_all[target_col])
+
 
 # -------------------------
 # helpers used in the pipeline but defined later
@@ -842,4 +915,41 @@ def create_forecast_index(last_train_date: pd.Timestamp, steps: int, freq: str =
     except Exception as e:
         print(f"Warning: could not create proper date index: {e}")
         # Fallback: числовой индекс
-        return pd.RangeIndex(start=0, stop=steps)
+        return pd.RangeIndex(start=0, stop=steps)
+
+
+def forecast_recursive(model, series, steps, freq='D'):
+    """Рекурсивная стратегия прогнозирования"""
+    predictions = []
+    current_series = series.copy()
+
+    for _ in range(steps):
+        if hasattr(model, 'predict'):
+            pred = model.predict(n_periods=1)
+        else:
+            pred = model.forecast(steps=1)
+        predictions.append(pred[0])
+        # Обновляем ряд для следующей итерации
+        current_series = pd.concat(
+            [current_series, pd.Series([pred[0]], index=[current_series.index[-1] + pd.Timedelta(days=1)])])
+
+    return np.array(predictions)
+
+
+def forecast_direct(train_series, test_features, model_factory, steps):
+    """Прямая стратегия - отдельная модель для каждого горизонта"""
+    predictions = []
+    for h in range(1, steps + 1):
+        # Создаем смещенную целевую переменную
+        y_h = train_series.shift(-h).dropna()
+        X_h = train_series.iloc[:len(y_h)]
+
+        # Обучаем модель для горизонта h
+        model = model_factory()
+        model.fit(X_h.values.reshape(-1, 1), y_h.values)
+
+        # Прогноз для горизонта h
+        pred = model.predict(train_series.values[-1:].reshape(1, -1))
+        predictions.append(pred[0])
+
+    return np.array(predictions)