machine_learning module
Developer entry point
Start with MLConfig, MLConfigurator, BaseWorkflow, PipelineBuilder, ModelFactory, and run_selector. See Developer Guide for overall architecture.
Core contracts
Workflows
Workflow classes wrap full ML pipelines: training, validation, and prediction.
Runners
Since 2026-05, K-fold compute loops live in the runners subpackage to simplify workflow classes while keeping the public API unchanged.
Note
KFoldWorkflow is the public entry point (legacy MachineLearningKFoldWorkflow remains as a deprecation subclass). It delegates fold training and aggregation to
habit.core.machine_learning.runners.kfold.KFoldRunner; persistence and plots use ReportWriter / ModelStore / PlotComposer.
Reporting components
habit.core.machine_learning.callbacks has been removed. Model storage, report writing, and visualization use explicit components in the reporting subpackage.