SchemaSampler#

class sdm.prior.tabular.SchemaSampler(task_type: Task, num_columns: HP[int] | int, cat_ratio: HP[float] | float, max_cardinality: HP[int] | int, num_classes: HP[int] | int | None = None, independent_cat_ratio: HP[float] | float = 0.0, max_correlated_cardinality: HP[int] | int = 9)#

Bases: object

Sample Schema instances.

Parameters:
  • task_type (Task) – Prediction task.

  • num_columns (HP[int] | int) – Feature count.

  • cat_ratio (HP[float] | float) – Categorical fraction, clipped to [0, 1].

  • max_cardinality (HP[int] | int) – Upper limit on category counts.

  • num_classes (HP[int] | int | None) – Class count, required for classification.

  • independent_cat_ratio (HP[float] | float) – Fraction of categorical features with independent cardinalities, clipped to [0, 1].

  • max_correlated_cardinality (HP[int] | int) – Additional cap on correlated cardinalities.

sample(generator: Generator | None = None, *, num_columns: int | None = None, num_classes: int | None = None) → Schema#

Sample a schema.

Parameters:
  • generator (Generator | None) – Pseudorandom generator.

  • num_columns (int | None) – Override the feature count for this draw.

  • num_classes (int | None) – Override the class count for a classification draw.

Return type:

Schema