monotonize#
- SuperGLM.monotonize(
- X: object,
- sample_weight: NDArray | None = None,
- offset: NDArray | None = None,
- *,
- n_grid: int = 500,
Repair post-fit shape-constrained spline terms after fitting.
This is a manual, post-fit repair step. It finds all spline features configured with
Constraint.postfit.*, applies the corresponding weighted shape projection to the fitted curve, and projects the repaired curve back to spline coefficients.Idempotent: calling twice does not re-repair already-repaired features.
- Parameters:
- Xpandas or eager Polars DataFrame
Training data (used to compute density-based grid weights).
- sample_weightarray-like, optional
Fitting weights used to weight the post-fit projection, read under the model’s declared
weight_semantics: replication counts under"frequency", precisions under"prior".- offsetarray-like, optional
Offset term (unused, reserved for deviance computation).
- n_gridint
Grid resolution for the shape repair (default 500).
- Returns:
- SuperGLM
The model (self), with post-fit shape repairs stored.