01 / IMPLEMENTATION & CONTRIBUTION
What the work involves
Donald extended Kieran Wood’s Momentum Transformer research with PyTorch modeling, execution environments, trade-delta policies, portfolio-aware agents, PPO corrections, and replay visualization.
Technical depth
Attention/LSTM financial sequence modeling, direct risk-objective optimization, regime features, portfolio-aware action projection, PPO/SAC training, GAE, observation normalization, transaction-cost accounting and execution adapters.
The project family
trading-momentum-transformer02 / RESULTS
What came out of it
Portfolio and execution-agent interfaces support threshold and reinforcement-learning policies. The work also corrected clipped-action log-probabilities and duplicate transaction penalties in the simulation/training path.
03 / SUPPORTING EVIDENCE
Follow the source
Implementation notes, project records, and supporting artifacts.
Source context & project scope
Original papers and their reported results belong to upstream authors. Names such as production execution layer denote code organization, not verified deployed profitability. No model or trading tests rerun in this audit.
Implemented base, threshold-signal and RL execution agents with trade projection.
SOURCE · 2026-09-17Adapter for reset, step, observation conversion and fills.
SOURCE · 2026-09-17Donald-attributed commits 0f35813, 7ec8ce7, 9aaa3dc and 38cca49 establish adaptation and corrected PPO/cost behavior.
SOURCE · 2026-09-17