FastICA<T>

Models & Types ยท AiDotNet.Preprocessing.DimensionalityReduction

Fast Independent Component Analysis (FastICA) for blind source separation.

Inheritance: TransformerBase<T, Matrix<T>, Matrix<T>> โ†’ FastICA<T>

For Beginners

Imagine you have recordings from multiple microphones in a room with multiple speakers. ICA can separate the individual speakers:

  • It finds components that are statistically independent
  • Unlike PCA which finds uncorrelated components, ICA finds truly independent ones
  • Works best when source signals are non-Gaussian (most real-world signals are)

How It Works

FastICA separates a multivariate signal into additive, independent non-Gaussian components. It is commonly used for blind source separation (e.g., separating mixed audio signals) and feature extraction.

The algorithm:

  1. Centers and whitens the data
  2. Uses fixed-point iteration to find independent components
  3. Each component is found by maximizing non-Gaussianity

Constructors

Constructor Summary
FastICA(int nComponents, ICAAlgorithm algorithm, ICAFunction fun, int maxIterations, double tolerance, bool whiten, int randomState, int[] columnIndices) Creates a new instance of FastICA.

Properties

Property Summary
ICAAlgorithm Algorithm { get; } Gets the algorithm type (parallel or deflation).
ICAFunction Fun { get; } Gets the non-linearity function used.
Matrix<T> Components { get; } Gets the unmixing matrix (components).
Matrix<T> Mixing { get; } Gets the mixing matrix.
bool SupportsInverseTransform { get; } Gets whether this transformer supports inverse transformation.
int NComponents { get; } Gets the number of independent components.

Methods

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