MaskLatent#

class sdm.prior.tabular.transform.MaskLatent(*, num_proxies: tuple[int, int] = (1, 3), columns: tuple[float, float] = (0.2, 0.6), prevalence: HP[float] | float | None = None, rate_off: HP[float] | None = None, rate_on: HP[float] | None = None)#

Bases: PriorTransform

Mask feature column blocks under an unobserved row factor.

Parameters:
  • num_proxies (tuple[int, int]) – Inclusive range for the number of proxy columns.

  • columns (tuple[float, float]) – Range for the fraction of non-proxy columns to mask.

  • prevalence (HP[float] | float | None) – Fraction of rows with the latent factor. Defaults to Uniform(0.1, 0.5).

  • rate_off (HP[float] | None) – Sampled missing rate per column block when the latent factor is inactive. The blocks share a correlated rate. Defaults to Uniform(0.0, 0.1).

  • rate_on (HP[float] | None) – Sampled missing rate per column block when the latent factor is active. The blocks share a correlated rate. Defaults to Uniform(0.3, 0.9).