Order‑book liquidity feature engineering + SHAP analysis (XGBoost) with Triple‑Barrier labels; reads OHLCV & depth from SQLite and saves plots to docs/images.

Logistic‑Normal Actor‑Critic für optimale Trade‑Ausführung in einem realistischen Limit‑Order‑Book‑Simulator (Noise/Tactical/Strategic); PyTorch‑Training inkl. TWAP/SL‑Baselines & Evaluation.

Deep‑Hedging in PyTorch (MCPG): europäische & amerikanische Optionen mit RSQP‑Risiko, GJR‑GARCH‑Pfade, IV‑Features und Chebyshev‑Pricing inkl. Baselines.

Rainflow‑aware DQN for battery dispatch (arbitrage + FR/reserves): Gymnasium env with safety layer, exact cycle‑based degradation, time features & stacked observations; SB3 training.