re#

superglm.re( column: str, *, levels: Any = None, unseen: Literal['population', 'error'] = 'population', missing: Literal['error'] = 'error', lambda_policy: LambdaPolicy | None = None, ) → BoundTerm#

Declare a random effect with a coefficient for every group level.

For example, re("broker") lets broker effects shrink toward the population value. Use fit_reml to estimate the variance component. This encoding does not drop a reference level.

Parameters:

columnstr: Name of the grouping column.
levelssequence, data column or categorical dtype, optional: Declare the allowed levels, including levels with no training rows.
unseen{“population”, “error”}, default=”population”: Prediction policy for unknown levels. "population" gives the random effect a contribution of zero on the predictor’s link scale.
missing{“error”}, default=”error”: Missing group labels raise an error.
lambda_policyLambdaPolicy, optional: Set the policy for the random effect’s smoothing penalty.

Returns:

BoundTerm: A random-effect declaration for the named column.