CoefficientInference.effective_parameters

property CoefficientInference.effective_parameters: float

Parameters the coefficient estimate actually spends.

Nominal for an unrestricted fit and strictly smaller under shrinkage, which is what makes it worth asking for uniformly: an information criterion that charges the nominal width penalizes a shrunk model for freedom it never used, and a caller comparing a shrunk fit against an unshrunk one has no way to notice unless both answer the same question.