Features

Feature specifications turn raw columns into model terms. Spline is the public factory for smooth terms; the concrete spline classes are what it returns. Categorical, ordered-categorical, numeric, polynomial and piecewise terms cover the rest of a rating structure; FactorSmooth and RandomEffect add credibility-style shrinkage; Constraint requests monotone or curvature constraints. The interaction types the constructor builds from a pair of columns are on the Internals page.

superglm.Spline	Create a spline feature spec.
superglm.PSpline	P-spline: B-spline basis with a discrete-difference penalty.
superglm.BSplineSmooth	B-spline smooth: B-spline basis with an integrated-derivative penalty.
superglm.NaturalSpline	Natural P-spline: f''=0 at boundaries, linear tails.
superglm.CubicRegressionSpline	CR spline: integrated f'' squared penalty + natural boundary constraints.
superglm.PolynomialRange	Pin a spline to a polynomial of degree (0-3) on [lo, hi].
superglm.n_knots_from_k	Convert basis dimension k to interior knot count.
superglm.Categorical	One-hot encoded categorical feature.
superglm.OrderedCategorical	Ordered categorical feature smoothed by a spline over its level values.
superglm.Numeric	A single continuous feature used as-is.
superglm.Piecewise	Continuous piecewise-linear feature on stated breakpoints.
superglm.Polynomial	Orthogonal polynomial feature (orthonormal in the training-weight measure).
superglm.FactorSmooth	A factor-by-P-spline interaction.
superglm.RandomEffect	All-level categorical effect with a REML-estimated variance component.
superglm.LevelGrouping	Mapping between original categorical levels and grouped levels.
superglm.collapse_levels	Build a LevelGrouping from data and grouping rules.
superglm.Constraint	The shape constraints a feature can carry, reached through Constraint.