Preprocessing quantiletransformeroptions

Development build. This page describes main, not a released package. The latest published Lodestar.Preprocessing is 0.1.0 — read its documentation.

Home › Preprocessing › Feature transforming

QuantileTransformerOptions

How many quantiles QuantileTransformer fits, and what it maps onto.

public sealed record QuantileTransformerOptions

Properties — QuantileCount is how many quantiles to fit; scikit-learn's n_quantiles, default 1000, clamped to the number of rows on both sides. Output is what the values are mapped onto; output_distribution, default QuantileOutput.Uniform.

Example — a coarse fit, which is a piecewise-linear approximation of the distribution.

using Lodestar.Preprocessing;

double[] skew = [1.0, 1.0, 2.0, 3.0, 5.0, 8.0, 13.0, 21.0, 34.0, 100.0];

QuantileTransformer coarse = QuantileTransformer.Fit(
    skew, 1, new QuantileTransformerOptions { QuantileCount = 5 });

double approximate = coarse.Transform([8.0])[0];   // => 0.53

Remarks — subsample is deliberately absent. At its own default of 10,000 the reference draws a subsample from numpy's generator, and two seeds were measured moving a quantile by 0.19 on twenty thousand rows. This transformer always reads every row, which is the reference's subsample=None — a narrower parameter set that is provable, rather than a wider one that is not.

random_state goes with it: it exists there only to seed that draw. copy is absent too, as everywhere in this package — a span goes in and a new array comes out, so there is nothing to copy in place.

Being a record, two option sets with the same two values are equal.

Applies to — net10.0, netstandard2.0.

See also — QuantileTransformer.Fit, QuantileOutput.