Project Genji: temporal model evaluation

An industry-aware equity research framework whose concrete implementation centers on ingestion, feature/label contracts, purged cross-validation and supervised model training.

PythonQlibXGBoostRidge regressionPurged cross-validationTushare

Built systemRank 08 of 13 in this field ↗

Why it sits here. Placed by the scope of implementation, the available contribution evidence, and the distinct technical capability it demonstrates.

03 / Somewhere between data and understanding.STUDY IN SPACE

01 / IMPLEMENTATION & CONTRIBUTION

What the work involves

Donald implemented feature and label contracts, time-aware cross-validation, supervised training, out-of-fold aggregation, and model persistence for industry-aware equity research.

Technical depth

Cross-sectional winsorization, industry normalization, multi-horizon training, purged/embargoed temporal splits, deterministic model comparison and structured prediction/evaluation contracts.

The project family

donald_trading_model

02 / RESULTS

What came out of it

Ridge and XGBoost training share purged and embargoed temporal splits, fold metrics, and aggregated out-of-fold predictions for structured model comparison.

03 / SUPPORTING EVIDENCE

Follow the source

Implementation notes, project records, and supporting artifacts.

Source context & project scope

Do not call all originally planned regime, QP portfolio, RL and backtest components complete. README plans and current implemented supervised-learning path must be distinguished.

README.md

Original system roadmap and data-ingestion scope; early status is stale.

SOURCE · 2026-09-17

train.py

Concrete CV splitter, trainers, fold metrics and OOF aggregation.

SOURCE · 2026-09-17

features.py

Industry normalization, winsorization and multi-family features.

SOURCE · 2026-09-17

Contribution history

b91cb79 and 2b15753 establish CV logging and multi-horizon training; 1a1db3d records determinism coverage.

SOURCE · 2026-09-17