LargeVis<T>
Models & Types ยท AiDotNet.Preprocessing.DimensionalityReduction
LargeVis for large-scale visualization and dimensionality reduction.
Inheritance: TransformerBase<T, Matrix<T>, Matrix<T>> โ LargeVis<T>
For Beginners
LargeVis is efficient for large datasets because:
- Uses approximate nearest neighbors for scalability
- Negative sampling reduces computation vs full graph
- Asynchronous updates enable parallel processing
- Layout preserves local neighborhood relationships
Use cases:
- Datasets with millions of points
- When t-SNE is too slow
- Interactive visualization systems
- Document and image embedding visualization
How It Works
LargeVis is designed for visualizing large-scale high-dimensional data. It builds a k-NN graph, computes edge weights based on shared neighbors, and uses asynchronous stochastic gradient descent with negative sampling for optimization.
The algorithm:
- Construct approximate k-NN graph using random projection trees
- Compute edge weights using shared neighbor similarity
- Initialize embedding with random projection or PCA
- Optimize using negative sampling SGD (similar to word2vec)
Constructors
| Constructor | Summary |
|---|---|
LargeVis(int nComponents, int nNeighbors, int nNegatives, int nIter, double learningRate, double gamma, int? randomState, int[] columnIndices) |
Creates a new instance of LargeVis. |
Properties
| Property | Summary |
|---|---|
bool SupportsInverseTransform { get; } |
Gets whether this transformer supports inverse transformation. |
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 |
|---|---|
string[] GetFeatureNamesOut(string[] inputFeatureNames) |
Gets the output feature names after transformation. |