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

Centres and scales a row-major sample matrix with the fitted statistics.

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, or copied unchanged when the scaler does not scale: a zero stays a zero, so nothing absent becomes stored.

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

Returns — a new array of the same length.

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, and InvalidOperationException when the scaler centres — fit it with WithCentring = false, since subtracting a centre would make every absent zero a stored value.

Example — subtract the median, then divide by the interquartile range.

using Lodestar.Preprocessing;

double[] samples = [1.0, 2.0, 3.0, 4.0, 5000.0];

RobustScaler scaler = RobustScaler.Fit(samples, featureCount: 1);

double[] robust = scaler.Transform(samples);

double smallest = robust[0];  // => -1
double middle = robust[2];    // => 0

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]);
RobustScaler scaler = RobustScaler.Fit(matrix);

CsrMatrix scaled = scaler.Transform(matrix);

int stored = scaled.NonZeroCount;   // => 2
int[] columns = scaled.ColumnIndices;   // same positions as the input

Remarks — subtract then divide, in that order, and the inverse multiplies then adds. Both steps are optional, so the order is part of the contract rather than an implementation detail: with centring off, nothing is subtracted and the division is applied to the raw value.

Never writes to the input.

Applies to — net10.0, netstandard2.0.

See also — RobustScaler.InverseTransform, RobustScalerOptions.