SplineRedundancyReport

class superglm.SplineRedundancyReport(

feature_name: str,

n_knots: int,

knot_locations: NDArray,

knot_spacing: NDArray,

support_mass: NDArray,

adjacent_basis_corr: NDArray,

coef_energy_penalized: NDArray,

effective_rank: float,

small_singular_values: NDArray = <factory>,

)

Bases: object

Redundancy diagnostics for one spline feature.

support_mass, adjacent_basis_corr, and effective_rank use replication mass under the frequency contract. Integer weights therefore match literal row replication. Under the prior contract the weights are validated but these geometric summaries use the physical rows rather than treating a precision as a row frequency; a Tweedie fit additionally requires them finite and strictly positive.