Shared standards, qualification evidence, documentation rules, and coordinated work management for the stable ML4T Python libraries.
| Library | Package | Responsibility |
|---|---|---|
| Data | ml4t-data |
Market data acquisition, validation, storage, and access |
| Engineer | ml4t-engineer |
Feature engineering, labeling, and dataset construction |
| Backtest | ml4t-backtest |
Event-driven simulation, execution, risk, and accounting |
| Specs | ml4t-specs |
Shared runtime-neutral lifecycle and trading contracts |
| Live | ml4t-live |
Paper and live execution using the shared contracts |
| Diagnostic | ml4t-diagnostic |
Statistical validation, splitters, evaluation, and diagnostics |
| Models | ml4t-models |
Model training, selection, persistence, and inference support |
This repository does not contain library source code and is not a runtime dependency. Start with the
ecosystem documentation or the tracked documents under
docs/standards/, status/, reviews/, and decisions/.
Keep ml4t-ecosystem/ beside the independent library checkouts in a plain parent directory.
Private development workspaces are sibling repositories; they are not part of this public
checkout. audit-workspaces uses that parent by default, while configuration, templates and
public evidence remain in this repository. Use --root to audit another library directory.
The shared contribution policy also covers Zipline Reloaded, Pyfolio Reloaded, Alphalens Reloaded and Empyrical Reloaded. Each retains its own repository, maintenance owner, native CI and documentation. Support covers bug, security and compatibility fixes and preserves docs. New features belong in current ML4T libraries. Backtest supplies Zipline migration guidance; Diagnostic supplies Pyfolio, Alphalens and Empyrical guidance. Migration claims require evidence for the specific workflow. Stable-release qualification below applies to the seven current libraries.
ML4T supports every stable CPython release from Python 3.12 through the latest stable version on Linux, macOS, and Windows. Each supported combination must pass installation, import, tests, type checking, and package build checks before release. The next CPython prerelease becomes a blocking compatibility target after beta 1 unless a library's required dependencies cannot yet pass it. A dependency-blocked library keeps a truthful Python upper bound and runs a visible non-blocking canary until its documented review triggers occur.
See Compatibility and Release qualification for the normative criteria.