GroupLasso

class superglm.GroupLasso(

lambda1: float | Literal['auto'] | None = None,

flavor: Flavor | None = None,

features: str | list[str] | None = None,

)

Bases: object

Group lasso penalty: lambda1 * sum_g(w_g * ||beta_g||_2).

Parameters:

lambda1float, {“auto”}, or None

Regularisation strength. None disables selection and "auto" explicitly requests calibration to 10% of lambda_max.

flavorFlavor or None

Optional modifier (e.g. Adaptive) that adjusts group weights based on an initial estimate.

prox_group(

bg: NDArray,

group: GroupSlice,

step: float,

) → NDArray

Block soft-thresholding for a single group.

prox(

beta: NDArray,

groups: list[GroupSlice],

step: float,

) → NDArray

Block soft-thresholding proximal operator.

eval(

beta: NDArray,

groups: list[GroupSlice],

) → float

Penalty value: lambda1 * sum_g(w_g * ||beta_g||_2).