QuantileTransform#

class sdm.processing.numerical.QuantileTransform(*, n_quantiles: int = 1000, subsample: int | None = 10000, output_distribution: Literal['uniform', 'normal'] = 'uniform')#

Bases: Processor, InvertibleMixin

Map numerical columns through their empirical quantiles.

QuantileTransform grids are capped by the number of fitted rows and, when subsample is set, by 20% of the subsample size to keep dense grids tractable.

The subsample rows are drawn from the generator passed to fit(); without one, they are drawn from the data device’s global generator.

Parameters:

  • n_quantiles (int) – Maximum number of quantiles to compute.

  • subsample (int | None) – Maximum number of rows to use for quantile computation.

  • output_distribution (Literal['uniform', 'normal']) – Distribution to map the empirical quantiles to.

Capabilities#

Handled Semantic Types

numerical

Requires Fitting

✅