Class TabularMixUp<T>

Namespace

AiDotNet.Augmentation.Tabular

Assembly

AiDotNet.dll

Applies MixUp augmentation to tabular data by linearly interpolating between samples.

public class TabularMixUp<T> : TabularMixingAugmenterBase<T>, ILabelMixingAugmentation<T, Matrix<T>>, IAugmentation<T, Matrix<T>>

Type Parameters

T

The numeric type for calculations.

Inheritance

object

AugmentationBase<T, Matrix<T>>

LabelMixingAugmentationBase<T, Matrix<T>>

TabularMixingAugmenterBase<T>

TabularMixUp<T>

Implements

ILabelMixingAugmentation<T, Matrix<T>>

IAugmentation<T, Matrix<T>>

Inherited Members

LabelMixingAugmentationBase<T, Matrix<T>>.OnLabelMixing

LabelMixingAugmentationBase<T, Matrix<T>>.LastMixingLambda

LabelMixingAugmentationBase<T, Matrix<T>>.Alpha

LabelMixingAugmentationBase<T, Matrix<T>>.SampleLambda(AugmentationContext<T>)

LabelMixingAugmentationBase<T, Matrix<T>>.RaiseLabelMixing(LabelMixingEventArgs<T>)

LabelMixingAugmentationBase<T, Matrix<T>>.GetParameters()

AugmentationBase<T, Matrix<T>>.NumOps

AugmentationBase<T, Matrix<T>>.OnAugmentationApplied

AugmentationBase<T, Matrix<T>>.Name

AugmentationBase<T, Matrix<T>>.Probability

AugmentationBase<T, Matrix<T>>.IsTrainingOnly

AugmentationBase<T, Matrix<T>>.IsEnabled

AugmentationBase<T, Matrix<T>>.Apply(Matrix<T>, AugmentationContext<T>)

AugmentationBase<T, Matrix<T>>.ApplyAugmentation(Matrix<T>, AugmentationContext<T>)

AugmentationBase<T, Matrix<T>>.GetParameters()

AugmentationBase<T, Matrix<T>>.RaiseAugmentationApplied(AugmentationContext<T>, bool)

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

For Beginners: MixUp creates new training samples by blending two existing samples together. If you have sample A and sample B, MixUp creates a new sample that's (λ × A) + ((1-λ) × B), where λ is randomly chosen. The labels are blended the same way.

Benefits:

• Regularizes the model by creating "virtual" training examples
• Encourages smooth decision boundaries
• Reduces overconfidence in predictions

When to use:

• Classification tasks with numerical features
• When you want to reduce overfitting
• When decision boundaries should be smooth

Reference: Zhang et al., "mixup: Beyond Empirical Risk Minimization" (2017)

Constructors

TabularMixUp(double, double)

Creates a new MixUp augmentation for tabular data.

public TabularMixUp(double alpha = 0.2, double probability = 0.5)

Parameters

alpha double

Alpha parameter for Beta distribution (default: 0.2). Smaller values keep samples closer to originals; larger values create more blending.

probability double

Probability of applying this augmentation (default: 0.5).

Methods

ApplyAugmentation(Matrix<T>, AugmentationContext<T>)

Implement this method to perform the actual augmentation.

protected override Matrix<T> ApplyAugmentation(Matrix<T> data, AugmentationContext<T> context)

Parameters

data Matrix<T>

The input data.

context AugmentationContext<T>

The augmentation context.

Returns

Matrix<T>

The augmented data.

MixWithLabels(Matrix<T>, Matrix<T>, Vector<T>, Vector<T>, AugmentationContext<T>)

Applies MixUp to two data matrices and their labels, returning mixed results.

public (Matrix<T> Data, Vector<T> Labels, double Lambda) MixWithLabels(Matrix<T> data1, Matrix<T> data2, Vector<T> labels1, Vector<T> labels2, AugmentationContext<T> context)

Parameters

data1 Matrix<T>

The first data matrix.

data2 Matrix<T>

The second data matrix.

labels1 Vector<T>

Labels for the first data matrix.

labels2 Vector<T>

Labels for the second data matrix.

context AugmentationContext<T>

The augmentation context.

Returns

(Matrix<T> Data, Vector<T> Labels, double Lambda)

Tuple of (mixed data, mixed labels, lambda value).