Package: PhysioML 0.4.2

PhysioML: Leakage-Aware Machine Learning for Physiological Time Series

Provides leakage-aware, reproducible feature transforms and reference deep-learning workflows for physiological time series stored in PhysioExperiment objects. Canonical catch22 characteristics are computed through the official R backend, while ROCKET and MiniRocket transforms use guarded optional Python backends. Guarded optional torch and luz backends provide case-aware datasets, compact neural architectures, training, and prediction. Explicit contracts preserve case, channel, feature, normalization, backend, and model provenance without claiming clinical validation. Named transfer-learning policies, leakage-safe CORAL alignment, and immutable hash-verified foundation-model caches provide bounded reuse across domains.

Authors:Yusuke Matsui [aut, cre]