Interface IPruningMask<T>

Namespace

AiDotNet.Interfaces

Assembly

AiDotNet.dll

Represents a binary mask for pruning weights in a neural network layer.

public interface IPruningMask<T>

Type Parameters

T

Numeric type for mask values

Remarks

A pruning mask is a binary matrix that determines which weights to keep (1) and which to remove (0) during model compression. It enables selective removal of network parameters while maintaining the ability to restore the network structure.

For Beginners: Think of a pruning mask as a stencil or template.

Imagine you're painting a picture and want to cover certain areas:

• The mask has holes (1s) where paint should go through (weights to keep)
• The mask is solid (0s) where paint should be blocked (weights to prune/remove)

In neural networks:

• A pruning mask helps you selectively remove less important connections
• This makes your model smaller and faster without losing too much accuracy
• The mask can be applied to weight matrices to zero out pruned weights

Properties

Pattern

Gets the sparsity pattern type.

SparsityPattern Pattern { get; }

Property Value

SparsityPattern

Shape

Gets the mask dimensions matching the weight matrix shape.

int[] Shape { get; }

Property Value

int[]

Methods

Apply(Matrix<T>)

Applies the mask to a weight matrix (element-wise multiplication).

Matrix<T> Apply(Matrix<T> weights)

Parameters

weights Matrix<T>

Weight matrix to prune

Returns

Matrix<T>

Pruned weights (zeros where mask is zero)

Apply(Tensor<T>)

Applies the mask to a weight tensor (for convolutional layers).

Tensor<T> Apply(Tensor<T> weights)

Parameters

weights Tensor<T>

Returns

Tensor<T>

Apply(Vector<T>)

Applies the mask to a vector.

Vector<T> Apply(Vector<T> weights)

Parameters

weights Vector<T>

Returns

Vector<T>

CombineWith(IPruningMask<T>)

Combines this mask with another mask (logical AND).

IPruningMask<T> CombineWith(IPruningMask<T> otherMask)

Parameters

otherMask IPruningMask<T>

Returns

IPruningMask<T>

GetKeptIndices()

Gets indices of non-zero (kept) elements.

int[] GetKeptIndices()

Returns

int[]

GetMaskData()

Gets the raw mask data as a flat array.

T[] GetMaskData()

Returns

T[]

GetPrunedIndices()

Gets indices of zero (pruned) elements.

int[] GetPrunedIndices()

Returns

int[]

GetSparsity()

Gets the sparsity ratio (proportion of zeros).

double GetSparsity()

Returns

double

Value between 0 (dense) and 1 (fully pruned)

Remarks

For Beginners: Sparsity measures how many weights have been removed. - 0.0 means no weights removed (0% sparse, 100% dense) - 0.5 means half the weights removed (50% sparse) - 0.9 means 90% of weights removed (90% sparse)

UpdateMask(Array)

Updates the mask with new N-D keep/prune decisions.

void UpdateMask(Array keepIndices)

Parameters

keepIndices Array

UpdateMask(bool[,])

Updates the mask based on new pruning criteria.

void UpdateMask(bool[,] keepIndices)

Parameters

keepIndices bool[,]

Indices of weights to keep (not prune)

UpdateMask(bool[])

Updates the mask with new keep/prune decisions.

void UpdateMask(bool[] keepIndices)

Parameters

keepIndices bool[]