Class L2Regularizer

Namespace

NeuralNetworks.Regularization

Assembly

NeuralNetworks.dll

Applies L2 regularization by adding a penalty proportional to the squared magnitude of model parameters.

public class L2Regularizer : BaseRegularizer, IRegularizer

Inheritance

object

BaseRegularizer

L2Regularizer

Implements

IRegularizer

Inherited Members

BaseRegularizer.ReportPenaltyInLoss

BaseRegularizer.ApplyGradient<T>(T, T)

BaseRegularizer.ReportPenalty<T>(T)

BaseRegularizer.UpdateParams<T>(T)

BaseRegularizer.UpdateParamsCore(Span<float>)

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

Remarks

Uses the L2 penalty form l2 * Σ(w^2), contributes 2 * l2 * w to each parameter gradient, and can optionally include the penalty term in the reported loss via the base regularizer configuration.

Constructors

L2Regularizer(float, bool)

Initializes a new instance of the L2Regularizer class with the specified L2 regularization factor.

public L2Regularizer(float l2, bool reportPenaltyInLoss = false)

Parameters

l2 float

Positive L2 regularization factor.

reportPenaltyInLoss bool

Indicates whether to include the regularization penalty in the reported loss.

Exceptions

ArgumentOutOfRangeException

Thrown when l2 is less than or equal to 0.

Methods

ApplyGradientCore(ReadOnlySpan<float>, Span<float>)

Adds the L2 regularization gradient contribution for each parameter element to the existing gradient values.

protected override void ApplyGradientCore(ReadOnlySpan<float> paramSpan, Span<float> paramGradientSpan)

Parameters

paramSpan ReadOnlySpan<float>

A read-only span containing the parameter values used to compute the L2 gradient term.

paramGradientSpan Span<float>

A span containing gradient values to update in place with the L2 contribution.

Remarks

For each index, adds 2 * _l2 * paramSpan[i] to paramGradientSpan[i].

CalculatePenaltyCore(ReadOnlySpan<float>)

Computes the L2 penalty as the scaled sum of squared parameter values.

protected override float CalculatePenaltyCore(ReadOnlySpan<float> paramSpan)

Parameters

paramSpan ReadOnlySpan<float>

A read-only span of parameter values to penalize.

Returns

float

The L2 coefficient multiplied by the sum of squares of all values in paramSpan.

ToString()

Returns a string that represents the current object.

public override string ToString()

Returns

string

A string that represents the current object.