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

Divides a row-major sample matrix by the fitted maxima.

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

public CsrMatrix Transform(CsrMatrix samples)

The second overload takes a CsrMatrix and returns a new one storing the same positions, each value divided by its column's Scale: a zero stays a zero, so nothing absent becomes stored. With clipping, a column stored twice in one row comes back stored once, its sum clamped, as the dense overload reads it.

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

Returns — a new array of the same length, inside [−1, 1] for any value the fit saw.

Exceptions — ArgumentException when samples holds no row, a partial one, or a non-finite value.

The sparse overload throws ArgumentNullException when samples is null, ArgumentException when it holds no row or its column count is not FeatureCount or it stores a non-finite value.

Example — a value beyond the fitted maximum leaves the unit interval, unless clipping is asked for.

using Lodestar.Preprocessing;

double[] seen = [0.0, 10.0];

MaxAbsScaler plain = MaxAbsScaler.Fit(seen, featureCount: 1);
MaxAbsScaler clipped = MaxAbsScaler.Fit(seen, 1, new MaxAbsScalerOptions { Clip = true });

double outside = plain.Transform([50.0])[0];    // => 5
double bounded = clipped.Transform([50.0])[0];  // => 1
double below = clipped.Transform([-50.0])[0];   // => -1

Example — the sparse overload, on a matrix whose second column stores nothing.

using Lodestar.Abstractions;
using Lodestar.Preprocessing;

// Three rows, two columns: 2 and -4 in the first column, nothing in the second.
var matrix = new CsrMatrix(3, 2, [2.0, -4.0], [0, 0], [0, 1, 1, 2]);
MaxAbsScaler scaler = MaxAbsScaler.Fit(matrix);

CsrMatrix scaled = scaler.Transform(matrix);

int stored = scaled.NonZeroCount;   // => 2
double second = scaled.Values[1];   // => -1

Remarks — never writes to the input. Clipping bounds this direction to [−1, 1] — the fitted range, which is fixed here rather than chosen as it is for MinMaxScaler — and leaves InverseTransform alone.

Applies to — net10.0, netstandard2.0.

See also — MaxAbsScalerOptions, MaxAbsScaler.InverseTransform.