Daily Audit 2026-04-25: logging violations, CatBoost optimization bug, global warning suppression, DataFrame fragmentation

[Copilot]

Description

Four new issues found during the 2026-04-25 codebase audit — not covered by any existing open issue — across logging, ML correctness, and performance categories.

Changes in this pull request

• Logging violations (base_pipeline.py, stacked_pipeline.py): traceback.print_exc() was called alongside logger.error() in exception handlers, violating the no-print() rule. Replaced with logger.exception() (logs at ERROR level + full traceback). Removed now-unused import traceback in both files.

• CatBoost optimization silently used LightGBM params (single_pipeline.py): _optimize_hyperparameters() had an if xgboost / else lightgbm structure — any model_type other than "xgboost" (including "catboost") ran LightGBM Optuna trials and passed those params to the model. Fixed with explicit if/elif/else; unsupported types now log a WARNING and fall back to defaults.

  # Before — catboost silently received LightGBM hyperparameters
  if self.model_type == "xgboost":
      optimizer.optimize_xgboost(); self._model_params = optimizer.best_xgb_params
  else:
      optimizer.optimize_lightgbm(); self._model_params = optimizer.best_lgb_params
  
  # After
  if self.model_type == "xgboost":
      optimizer.optimize_xgboost(); self._model_params = optimizer.best_xgb_params
  elif self.model_type == "lightgbm":
      optimizer.optimize_lightgbm(); self._model_params = optimizer.best_lgb_params
  else:
      logger.warning("Hyperparameter optimisation is not supported for model_type='%s'. ...", self.model_type)

• Global warnings.filterwarnings('ignore') in main.py: Separate untracked occurrence of the pattern already captured in Performance: warnings.filterwarnings('ignore') suppresses all warnings globally #103 for optimizer.py. Removed.

• DataFrame fragmentation (feature_engineering.py, base_pipeline.py): add_all_technical_indicators() inserts ~130 columns one-by-one, causing pandas PerformanceWarning: DataFrame is highly fragmented at lines 282, 289, and 293. Added self.df = self.df.copy() midway to consolidate blocks before late indicators. Also refactored _add_stock_features() from 6 individual column assignments to a single pd.concat. Test-suite PerformanceWarnings dropped from 1503 → 29.