Preprocessing standardscaler

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

StandardScaler

Centres each feature on its mean and scales it to unit variance.

public sealed class StandardScaler

Properties — FeatureCount is how many values a row carries, and SampleCount how many rows the scaler was fitted on (scikit-learn's n_samples_seen_). Mean, Variance and Scale are the fitted statistics, each nullable, because the reference reports None for the ones a given option pair does not compute — see the table below.

Example — fit on a matrix, then standardise it.

using Lodestar.Preprocessing;

// Row-major: two features per row, three rows. The second is constant.
double[] samples = [1.0, 10.0, 2.0, 10.0, 4.0, 10.0];

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

double firstMean = scaler.Mean![0];      // => 2.3333333333333335
double constantScale = scaler.Scale![1]; // => 1

double[] standardised = scaler.Transform(samples);
double first = standardised[0];          // => -1.069…

Remarks — which statistics exist depends on both options, and not in the obvious way. This is scikit-learn's behaviour, reproduced rather than tidied:

WithMean	WithStd	Mean	Variance	Scale
true	true	fitted	fitted	fitted
false	true	fitted	fitted	fitted
true	false	fitted	null	null
false	false	null	null	null

Turning centring off still computes a mean; only turning both steps off drops it. A scaler that reports a Mean is therefore not necessarily a scaler that subtracts one — what StandardScaler.Transform does is decided by the options, not by which statistics came back.

The second feature above is constant, and its Scale is 1 rather than 0. The rule behind that is not variance == 0, and it is worth reading before trusting a value near the noise floor: StandardScaler.Fit has it.

Applies to — net10.0, netstandard2.0.

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

Members

Member	What it does
StandardScaler.Fit	Fits a scaler on a row-major sample matrix.
StandardScaler.InverseTransform	Undoes Transform, returning values on the original scale.
StandardScaler.PartialFit	Folds another batch into the fitted statistics.
StandardScaler.Transform	Standardises a matrix with the fitted statistics.