Class HybridHuffmanClusteringCompression<T>

Namespace

AiDotNet.ModelCompression

Assembly

AiDotNet.dll

public class HybridHuffmanClusteringCompression<T> : ModelCompressionBase<T>, IModelCompressionStrategy<T>

Type Parameters

T

Inheritance

object

ModelCompressionBase<T>

HybridHuffmanClusteringCompression<T>

Implements

IModelCompressionStrategy<T>

Inherited Members

ModelCompressionBase<T>.NumOps

ModelCompressionBase<T>.CalculateCompressionRatio(long, long)

ModelCompressionBase<T>.CompressMatrix(Matrix<T>)

ModelCompressionBase<T>.DecompressMatrix(Matrix<T>, ICompressionMetadata<T>)

ModelCompressionBase<T>.GetCompressedSize(Matrix<T>, ICompressionMetadata<T>)

ModelCompressionBase<T>.CompressTensor(Tensor<T>)

ModelCompressionBase<T>.DecompressTensor(Tensor<T>, ICompressionMetadata<T>)

ModelCompressionBase<T>.GetCompressedSize(Tensor<T>, ICompressionMetadata<T>)

ModelCompressionBase<T>.GetElementSize()

ModelCompressionBase<T>.MatrixToVector(Matrix<T>)

ModelCompressionBase<T>.VectorToMatrix(Vector<T>, int, int)

ModelCompressionBase<T>.TensorToVector(Tensor<T>)

ModelCompressionBase<T>.VectorToTensor(Vector<T>, int[])

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Constructors

HybridHuffmanClusteringCompression(int, int, double, int, int?)

Initializes a new instance of the HybridHuffmanClusteringCompression class.

public HybridHuffmanClusteringCompression(int numClusters = 256, int maxIterations = 100, double tolerance = 1E-06, int huffmanPrecision = 0, int? randomSeed = null)

Parameters

numClusters int

Number of clusters for weight clustering (default: 256).

maxIterations int

Maximum K-means iterations (default: 100).

tolerance double

K-means convergence tolerance (default: 1e-6).

huffmanPrecision int

Precision for Huffman encoding (default: 0 for cluster indices).

randomSeed int?

Random seed for reproducibility.

Remarks

For Beginners: These parameters control both compression stages.

Clustering parameters:

• numClusters: How many groups to create (256 = 8-bit, very common)

  • More clusters = better quality, less compression
  • Fewer clusters = more compression, lower quality

• maxIterations/tolerance: How hard to work on finding optimal clusters

  • Defaults are usually fine

Huffman parameters:

• huffmanPrecision: Set to 0 for cluster indices (they're already integers)
  • Higher values only matter for floating-point data

The magic happens when these work together:

• Clustering creates patterns
• Huffman exploits those patterns
• Total compression is better than either alone

Methods

Compress(Vector<T>)

Compresses weights using clustering followed by Huffman encoding.

public override (Vector<T> compressedWeights, ICompressionMetadata<T> metadata) Compress(Vector<T> weights)

Parameters

weights Vector<T>

The original model weights.

Returns

(Vector<T> compressedWeights, ICompressionMetadata<T> metadata)

Compressed weights and hybrid metadata.

Decompress(Vector<T>, ICompressionMetadata<T>)

Decompresses weights by reversing Huffman encoding then clustering.

public override Vector<T> Decompress(Vector<T> compressedWeights, ICompressionMetadata<T> metadata)

Parameters

compressedWeights Vector<T>

The compressed weights.

metadata ICompressionMetadata<T>

The hybrid compression metadata.

Returns

Vector<T>

The decompressed weights.

GetCompressedSize(Vector<T>, ICompressionMetadata<T>)

Gets the total compressed size from both compression stages.

public override long GetCompressedSize(Vector<T> compressedWeights, ICompressionMetadata<T> metadata)

Parameters

compressedWeights Vector<T>

metadata ICompressionMetadata<T>

Returns

long