Preprocessing powertransformer inversetransform

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

PowerTransformer.InverseTransform

Undoes Transform.

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

Parameters — samples is a matrix this transformer produced, row-major, FeatureCount values per row.

Returns — a new matrix of the same shape, each value read back through the standardisation and then the power.

Exceptions — ArgumentException when samples holds a partial row or a non-finite value.

Example — a value through and back.

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 transformer = PowerTransformer.Fit(income, 1);

double[] there = transformer.Transform([35.0]);

double back = Math.Round(transformer.InverseTransform(there)[0], 4);   // => 35

Remarks — unlike the two neighbours in this section, this round trip is genuine. The power is a monotone function with an exact inverse, so nothing is discarded on the way out and the value returns to within rounding — which is what makes this transformer usable on a target variable, a model fitted on the transformed scale and its predictions read back on the original one.

The one place it does not is where the power cannot be undone. Under PowerMethod.BoxCox the transformed scale is bounded on one side — below by -1/λ for a positive exponent, above by it for a negative one — because no positive value maps past that asymptote. A value outside comes back as NaN, which is what the reference returns there too, rather than a number the family never produced.

Applies to — net10.0, netstandard2.0.

See also — PowerTransformer.Transform, PowerTransformer.