Isomap<T>
Models & Types · AiDotNet.Preprocessing.DimensionalityReduction
Isomap (Isometric Mapping) for nonlinear dimensionality reduction.
Inheritance: TransformerBase<T, Matrix<T>, Matrix<T>> → Isomap<T>
For Beginners
Isomap “unrolls” curved data:
- Imagine data lying on a Swiss roll (curved surface)
- Regular PCA can’t flatten it properly
- Isomap finds distances along the surface, not through the air
- Result: Points that are nearby on the surface stay nearby
How It Works
Isomap extends classical MDS by estimating geodesic distances along the data manifold instead of using Euclidean distances. It builds a neighborhood graph and computes shortest paths to estimate geodesic distances.
The algorithm:
- Build a k-nearest neighbors graph
- Compute shortest path distances (Floyd-Warshall or Dijkstra)
- Apply MDS to the geodesic distance matrix
Constructors
| Constructor | Summary |
|---|---|
Isomap(int nComponents, int nNeighbors, IsomapNeighborAlgorithm neighborAlgorithm, IsomapPathAlgorithm pathAlgorithm, int[] columnIndices) |
Creates a new instance of Isomap. |
Properties
| Property | Summary |
|---|---|
Matrix<T> Embedding { get; } |
Gets the embedding result. |
Matrix<T> GeodesicDistances { get; } |
Gets the geodesic distance matrix. |
bool SupportsInverseTransform { get; } |
Gets whether this transformer supports inverse transformation. |
int NComponents { get; } |
Gets the number of components. |
int NNeighbors { get; } |
Gets the number of neighbors for graph construction. |
Methods
| Method | Summary |
|---|---|
string[] GetFeatureNamesOut(string[] inputFeatureNames) |
Gets the output feature names after transformation. |