fit#

SuperLSS.fit( X: object | EagerFrame, y: NDArray, sample_weight: NDArray | None = None, offsets: Mapping[str, NDArray] | None = None, *, lambdas: Mapping[str, float] | None = None, max_inner_iter: int = 100, inner_tol: float = 1e-07, retain_rows: bool = True, ) → SuperLSS#

Fit coefficients while holding smoothing parameters fixed.

Parameters:

XDataFrame or EagerFrame: Input columns named in the predictor declarations.
yarray-like of shape (n_observations,): Response values in the family’s support, in the same row order as X.
sample_weightarray-like, optional: One weight per row, interpreted using weight_semantics.
offsetsmapping of str to array-like, optional: Known additions to predictors on their link scales. Keys are family parameter names, such as "mean" for Tweedie.
lambdasmapping of str to float, optional: Smoothing strengths keyed by fully qualified penalty names, such as "location:age#wiggle" for a Gaussian smooth.
max_inner_iterint, default=100: Maximum number of coefficient iterations.
inner_tolfloat, default=1e-7: Coefficient convergence tolerance.
retain_rowsbool, default=True: Retain fitted row arrays for training diagnostics.

Returns:

SuperLSS: This model, with its fitted state replaced by the new fit.