LambdaPolicy

class superglm.LambdaPolicy(mode: Literal['estimate', 'fixed'], value: float | None = None)

Bases: object

Per-penalty-component lambda control for REML.

Two modes: - estimate: lambda optimised by REML, initialised from lambda2_init. - fixed(value): lambda held at value throughout fitting.

off() is sugar for fixed(0.0) — the penalty component contributes nothing to the penalty matrix.