RandomProjection<T>

Models & Types · AiDotNet.Preprocessing.DimensionalityReduction

Random Projection for dimensionality reduction.

Inheritance: TransformerBase<T, Matrix<T>, Matrix<T>> → RandomProjection<T>

For Beginners

Random projection is surprisingly effective:

  • It’s very fast (just matrix multiplication)
  • It preserves distances approximately (guaranteed by math!)
  • Great for preprocessing before other algorithms
  • Works well for very high-dimensional data

Use cases:

  • Speeding up distance-based algorithms (kNN, clustering)
  • Reducing memory for large datasets
  • Preprocessing before other dimensionality reduction

How It Works

Random projection reduces dimensionality by projecting data onto a random subspace. Despite its simplicity, it has strong theoretical guarantees via the Johnson-Lindenstrauss lemma: pairwise distances are approximately preserved with high probability.

Two projection types are supported:

  • Gaussian: Random matrix with entries from N(0, 1/k)
  • Sparse: Random matrix with mostly zeros (faster, memory efficient)

Constructors

Constructor Summary
RandomProjection(int? nComponents, double? eps, RandomProjectionType projectionType, double density, int? randomState, int[] columnIndices) Creates a new instance of RandomProjection.

Properties

Property Summary
Matrix<T> ProjectionMatrix { get; } Gets the projection matrix.
RandomProjectionType ProjectionType { get; } Gets the projection type.
bool SupportsInverseTransform { get; } Gets whether this transformer supports inverse transformation.
int NComponents { get; } Gets the number of components.

Methods

Method Summary
string[] GetFeatureNamesOut(string[] inputFeatureNames) Gets the output feature names after transformation.