Class HuberLoss<T>

Namespace

AiDotNet.LossFunctions

Assembly

AiDotNet.dll

Implements the Huber loss function, which combines properties of both MSE and MAE.

public class HuberLoss<T> : LossFunctionBase<T>, ILossFunction<T>

Type Parameters

T

The numeric type used for calculations (e.g., float, double).

Inheritance

object

LossFunctionBase<T>

HuberLoss<T>

Implements

ILossFunction<T>

Inherited Members

LossFunctionBase<T>.NumOps

LossFunctionBase<T>.Engine

LossFunctionBase<T>.ValidateVectorLengths(Vector<T>, Vector<T>)

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Extension Methods

LossFunctionExtensions.ComputeGradient<T>(ILossFunction<T>, Tensor<T>, Tensor<T>)

LossFunctionExtensions.ComputeLoss<T>(ILossFunction<T>, Tensor<T>, Tensor<T>)

Remarks

For Beginners: Huber loss combines the best properties of Mean Squared Error and Mean Absolute Error.

The formula is:

• For errors smaller than delta: 0.5 * error²
• For errors larger than delta: delta * (|error| - 0.5 * delta)

Where "error" is the difference between predicted and actual values.

Key properties:

• For small errors, it behaves like MSE (quadratic/squared behavior)
• For large errors, it behaves like MAE (linear behavior)
• Less sensitive to outliers than MSE, but still provides smooth gradients
• The delta parameter controls the transition point between quadratic and linear regions

Huber loss is ideal for regression problems where:

• You want to balance between MSE and MAE
• Your data might contain outliers
• You need stable gradients for learning

The delta parameter lets you control the definition of an "outlier" - errors larger than delta are treated as outliers and handled using the more robust linear function.

Constructors

HuberLoss(double)

Initializes a new instance of the HuberLoss class with the specified delta.

public HuberLoss(double delta = 1)

Parameters

delta double

The threshold parameter that controls the transition point. Default is 1.0.

Methods

CalculateDerivative(Vector<T>, Vector<T>)

Calculates the derivative of the Huber loss function.

public override Vector<T> CalculateDerivative(Vector<T> predicted, Vector<T> actual)

Parameters

predicted Vector<T>

The predicted values from the model.

actual Vector<T>

The actual (target) values.

Returns

Vector<T>

A vector containing the derivatives of Huber loss for each prediction.

CalculateLoss(Vector<T>, Vector<T>)

Calculates the Huber loss between predicted and actual values.

public override T CalculateLoss(Vector<T> predicted, Vector<T> actual)

Parameters

predicted Vector<T>

The predicted values from the model.

actual Vector<T>

The actual (target) values.

Returns

T

The Huber loss value.

CalculateLossAndGradientGpu(IGpuTensor<T>, IGpuTensor<T>)

Calculates both Huber loss and gradient on GPU in a single efficient pass.

public override (T Loss, IGpuTensor<T> Gradient) CalculateLossAndGradientGpu(IGpuTensor<T> predicted, IGpuTensor<T> actual)

Parameters

predicted IGpuTensor<T>

The predicted GPU tensor from the model.

actual IGpuTensor<T>

The actual (target) GPU tensor.

Returns

(T Loss, IGpuTensor<T> Gradient)

A tuple containing the loss value and gradient tensor.

Remarks

Uses the SmoothL1 kernel which is equivalent to Huber loss with beta = delta.