TriMAP<T>

Models & Types ยท AiDotNet.Preprocessing.DimensionalityReduction

TriMAP: Large-scale Dimensionality Reduction Using Triplets.

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

For Beginners

TriMAP preserves distance relationships using triplets:

  • A triplet (i, j, k) means point i is closer to j than to k
  • Optimization ensures these relationships hold in the embedding
  • More accurate than t-SNE/UMAP for many datasets
  • Less sensitive to parameter tuning

Use cases:

  • Large-scale visualization (millions of points)
  • When t-SNE/UMAP produces poor results
  • When you need faithful global structure
  • Complex datasets with hierarchical structure

How It Works

TriMAP is a dimensionality reduction method that uses triplet constraints to preserve both local and global structure. It outperforms t-SNE and UMAP on many datasets while being more robust to hyperparameter choices.

The algorithm:

  1. Generate triplets (anchor, positive, negative) based on distance relationships
  2. Initialize embedding using PCA
  3. Optimize using triplet loss to preserve distance rankings
  4. Weight triplets to balance local and global structure

Constructors

Constructor Summary
TriMAP(int nComponents, int nInliers, int nOutliers, int nRandom, int nIter, double learningRate, int? randomState, int[] columnIndices) Creates a new instance of TriMAP.

Properties

Property Summary
Matrix<T> Embedding { get; } Gets the embedding result.
bool SupportsInverseTransform { get; } Gets whether this transformer supports inverse transformation.
int NComponents { get; } Gets the number of components (dimensions).

Methods

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