Build

SuperGLM is built from a family, a feature specification and a penalty policy, with every constructor argument described on its page. Once the model exists, the properties below hold that configuration, and the trailing-underscore attributes are filled in by the fit.

superglm.SuperGLM

Penalised generalised linear model with splines, group penalties, and REML.

Configuration and fitted attributes

family, link, penalty and selection_penalty hold the configuration and can be assigned to change it before the next fit; lambda2 is the fixed smoothing penalty the constructor takes as spline_penalty, assignable the same way; and features is read-only. The trailing-underscore attributes follow the scikit-learn convention and are resolved by the latest successful fit: selection_penalty_, distribution_ and theta_, the NB2 dispersion.

family	Defensive copy of configured family intent.
link	Defensive copy of configured link intent.
features	Defensive fitted/configured feature-spec view.
penalty	Defensive copy of the configured penalty template.
lambda2	Defensive copy of configured smoothing penalties.
selection_penalty	Configured selection-penalty intent.
selection_penalty_	Resolved selection penalty from the latest successful fit.
distribution_	Resolved distribution from the latest successful fit.
theta_	Resolved NB2 dispersion parameter from the latest successful fit.