Skip to content

Machine Learning Overview

This section documents the machine learning lifecycle of the Time2Bet system: from problem formulation to model registration and promotion.

The primary goals of the ML layer are: - reproducible and traceable training, - leakage-safe validation, - explicit model contracts, - safe and auditable deployment into production.

ML design principles

  • Reproducibility by default Training results are deterministic with respect to data, code, and configuration.

  • Validation over optimization Correct validation strategy is prioritized over marginal metric gains.

  • Explicit contracts Models expose well-defined input/output schemas.

  • Separation of concerns Feature engineering, training logic, and serving are decoupled via artifacts.

ML lifecycle (high level)

flowchart LR A[Versioned Dataset] --> B[Feature Engineering] B --> C[Train / Validate] C --> D[Evaluate] D --> E[MLflow Tracking] E --> F[Model Registry] F --> G[Serving]