API reference

One page per part of the library, in dependency order: models are built from layers, layers from functions. Each page starts with what the objects are for and a summary table; the detailed entries follow. For the reasoning behind them (geometries, gradient paths, parametrizations), see the User guide.

Page	Contents
Models	SPDnet, RResNet, GBWBNRResNet: complete classifiers
Layers	BiMap, ReEig, ReEigBias, LogEig, Vec, Vech; covariance estimation layers
Batch normalization	Riemannian batch normalization (mean only, mean + scalar variance, GBWBN)
Residual blocks	Spectral vector field and residual block of the residual networks
Parametrizations	SPD, Stiefel and positive-scalar parametrizations of the parameters
SPD linear algebra	Matrix functions through eigendecomposition, congruences, vectorizations
Geometries	Geodesics, means, dispersions, exp/log maps of the five geometries
Covariance estimators	Sample covariance and robust M-estimators of scatter
Utilities	Scalar maps, Stiefel projections, random SPD/Stiefel matrices