UMAP<T>

Models & Types · AiDotNet.Preprocessing.DimensionalityReduction

Uniform Manifold Approximation and Projection for dimensionality reduction.

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

For Beginners

UMAP creates visualizations similar to t-SNE but:

  • It’s faster, especially for large datasets
  • Distances between clusters are more meaningful
  • You can transform new data points without refitting
  • Great for both visualization AND as a preprocessing step for ML

Example use cases:

  • Visualizing high-dimensional data (gene expression, embeddings)
  • Preprocessing features for classification
  • Clustering analysis
  • Anomaly detection

How It Works

UMAP is a nonlinear dimensionality reduction technique that constructs a high-dimensional graph representation and optimizes a low-dimensional graph to be as structurally similar as possible. It is based on Riemannian geometry and algebraic topology.

Key advantages over t-SNE:

  • Much faster (scales better to large datasets)
  • Preserves more global structure
  • Supports out-of-sample transformation
  • More deterministic results

Constructors

Constructor Summary
UMAP(int nComponents, int nNeighbors, double minDist, double spread, UMAPMetric metric, int nEpochs, double learningRate, double negativeSampleRate, double localConnectivity, double repulsionStrength, int? randomState, int[] columnIndices) Creates a new instance of UMAP.

Properties

Property Summary
UMAPMetric Metric { get; } Gets the distance metric.
bool SupportsInverseTransform { get; } Gets whether this transformer supports inverse transformation.
double MinDist { get; } Gets the minimum distance parameter.
double[] Embedding { get; } Gets the embedding result.
int NComponents { get; } Gets the number of components (dimensions).
int NNeighbors { get; } Gets the number of neighbors.

Methods

Method Summary
Matrix<T> GetEmbedding() Gets the embedding computed during Fit for the training data.
string[] GetFeatureNamesOut(string[] inputFeatureNames) Gets the output feature names after transformation.