Development build. This page describes main, not a released package. The latest published Lodestar.Preprocessing is 0.2.0 — read its documentation.

Home › Preprocessing › Feature transforming

Raises each feature to its fitted power.

public double[] Transform(ReadOnlySpan<double> samples)

Parameters — samples is the matrix to transform, row-major, FeatureCount values per row.

Returns — a new matrix of the same shape, each feature raised to its fitted exponent and, when PowerTransformerOptions.Standardize is on, centred and scaled afterwards.

Exceptions — ArgumentException when samples holds a partial row or a non-finite value; when the family is PowerMethod.BoxCox and a value is not strictly positive; or, standardised, when the power overflows a value, as the reference's StandardScaler refuses the infinity — unstandardised, the infinity is returned (#1230).

Example — with and without the standardisation step.

using Lodestar.Preprocessing;

double[] income = [22.0, 25.0, 28.0, 31.0, 35.0, 42.0, 55.0, 78.0, 120.0, 260.0];

PowerTransformer standardised = PowerTransformer.Fit(income, 1);
PowerTransformer raw = PowerTransformer.Fit(
    income, 1, new PowerTransformerOptions { Standardize = false });

double centred = Math.Round(standardised.Transform([22.0])[0], 4);   // => -1.4543
double powered = Math.Round(raw.Transform([22.0])[0], 4);            // => 1.1403

Remarks — a value outside the fitted range is transformed, not clamped. The exponent is a formula, so a later row larger than anything fitted comes out larger than anything fitted, which is the behaviour QuantileTransformer cannot offer. What the fit decided was the exponent and, under standardisation, the mean and deviation of the transformed column; none of the three bounds a later value.

The standardisation is a StandardScaler's, as the reference's is. It is fitted after the power, on the fitted rows — so a fitted column comes out with mean 0 and deviation 1, and a later row does not necessarily — except on a constant Yeo-Johnson feature, which the reference leaves untransformed at fit and so standardises as it was (#1230).

Applies to — net10.0, netstandard2.0.

See also — PowerTransformer.InverseTransform, PowerTransformerOptions.