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 scaling

MaxAbsScaler

Divides each feature by its largest absolute value.

public sealed class MaxAbsScaler

Properties — FeatureCount and SampleCount are the shape it was fitted on. MaximumAbsolute is each feature's largest absolute value and Scale is what Transform divides by — the same numbers, except on a feature the near-constant floor caught.

Example — two features, one of them negative throughout.

using Lodestar.Preprocessing;

double[] samples = [2.0, -4.0, 1.0, -8.0];

MaxAbsScaler scaler = MaxAbsScaler.Fit(samples, featureCount: 2);

double first = scaler.MaximumAbsolute[0];  // => 2
double second = scaler.Scale[1];           // => 8

double scaled = scaler.Transform(samples)[1];  // => -0.5

Remarks — it never subtracts, so a zero stays a zero. That is the whole reason to prefer it to StandardScaler on data whose zeros are meaningful: centring turns every zero into a non-zero, which is why scikit-learn refuses to centre a sparse matrix at all.

The largest absolute value is not the largest value: the second feature above runs from −8 to −4 and is divided by 8.

Applies to — net10.0, netstandard2.0.

See also — MaxAbsScalerOptions, the feature scaling index, the Python equivalence table.

Members

Member	What it does
MaxAbsScaler.Fit	Fits a scaler on a row-major sample matrix.
MaxAbsScaler.InverseTransform	Undoes Transform, and never clips.
MaxAbsScaler.PartialFit	Folds another batch into the fitted statistics.
MaxAbsScaler.Transform	Divides a matrix by the fitted maxima.