StandardScaler

classin packageklyn.math.preprocessing

public class StandardScaler<T extends AbstractFloat = Float>

StandardScaler

Learns per-feature training statistics and reuses them on new observations. Inputs are two-dimensional [samples, features] floating tensors. Centering and population-standard-deviation scaling are enabled by default. A constant feature uses scale one, so its training values become zero when centered. Statistics use Double; output storage keeps T, including physical FP16. Instances must not be fitted concurrently with fit/transform calls.

Example

scaler = StandardScaler<Float>()

xTrain = scaler.fitTransform(xTrain, inPlace=true)

xTest = scaler.transform(xTest, inPlace=true)

Properties

Modifier and Type	Member	Description
public readonly property	isFitted `isFitted as Boolean`	Whether fit has completed successfully.
public readonly property	mean `mean as IList<Double>`	Immutable applied offsets, cached until the next successful fit.
public readonly property	scale `scale as IList<Double>`	Immutable applied divisors; one for constants or disabled scaling.
public readonly property	withMean `withMean as Boolean`	Whether to subtract training means.
public readonly property	withStd `withStd as Boolean`	Whether to divide by training standard deviations.

Constructors

Modifier and Type	Member	Description
public	StandardScaler `StandardScaler(withMean as Boolean = true, withStd as Boolean = true)`	Configures centering and scaling independently, without fitting data.

Methods

Modifier and Type	Member	Description
public	fit `fit(notnull data as NDArray<T>) as StandardScaler<T>`	Fits on training observations only.
public	fitTransform `fitTransform(notnull data as NDArray<T>, inPlace as Boolean = false) as NDArray<T>`	Fits training statistics, then applies them using transform's copy policy.
public	inverseTransform `inverseTransform(notnull data as NDArray<T>, inPlace as Boolean = false) as NDArray<T>`	Applies x * trainingScale + trainingMean with the same copy policy.
public	transform `transform(notnull data as NDArray<T>, inPlace as Boolean = false) as NDArray<T>`	Applies (x - trainingMean) / trainingScale using one fused buffer pass.