Coarsen#

class sdm.prior.tabular.transform.Coarsen(*, columns: HP[float] | float | None = None, num_levels: HP[int] | None = None, step: HP[float] | None = None, grid_probability: float = 0.5, categorical: bool = False, standardize: bool = False)#

Bases: PriorTransform

Reduce selected feature columns to a few ordered levels.

Parameters:
  • columns (HP[float] | float | None) – Fraction of eligible columns to coarsen per dataset. A value above one selects all columns. Defaults to Uniform(0.3, 1.3).

  • num_levels (HP[int] | None) – Sampled number of output levels per column. The columns of one dataset share a correlated distribution. Defaults to LogUniformInt(2, 8).

  • step (HP[float] | None) – Sampled regular-grid spacing per column. The columns of one dataset share a correlated distribution. Defaults to LogUniform(0.25, 2.0).

  • grid_probability (float) – Probability of using a regular grid.

  • categorical (bool) – Whether categorical columns are eligible.

  • standardize (bool) – Whether to standardize coarsened values.