Overview TSNE (t-Distributed Stochastic Neighbor Embedding) is a non-linear dimensionality reduction technique particularly well-suited for visualizing high-dimensional data in 2D or 3D space.
Constructor new TSNE ( options ?: TSNEOptions )
Configuration options for TSNE Dimension of the embedded space (typically 2 or 3 for visualization).
Related to the number of nearest neighbors. Larger datasets require larger perplexity. Typical values: 5-50.
Learning rate for gradient descent. Typical values: 10-1000.
Maximum number of iterations for optimization (minimum 250).
Seed for random number generator for reproducible results.
Methods fit Fit the TSNE model on training data.
Training data of shape [nSamples, nFeatures]
Returns: The fitted TSNE instancefitTransform ( X : Matrix ): Matrix
Fit the model and return the embedding.
Training data of shape [nSamples, nFeatures]
Returns: Embedded data of shape [nSamples, nComponents]Properties Embedded coordinates in low-dimensional space, shape [nSamples, nComponents]
Number of features seen during fit
Kullback-Leibler divergence after optimization
Example import { TSNE } from '@elucidate/elucidate' ; // High-dimensional data for visualization const data = [ [ 2.5 , 3.2 , 1.8 , 4.1 ], [ 2.3 , 3.0 , 1.7 , 3.9 ], [ 5.1 , 6.2 , 4.3 , 7.8 ], [ 5.3 , 6.5 , 4.5 , 8.1 ], [ 1.2 , 1.5 , 0.8 , 1.9 ], [ 1.0 , 1.3 , 0.7 , 1.7 ], ]; const tsne = new TSNE ({ nComponents: 2 , perplexity: 3 , learningRate: 200 , maxIter: 1000 , randomState: 42 }); const embedding = tsne . fitTransform ( data ); console . log ( '2D embedding:' , embedding ); console . log ( 'KL divergence:' , tsne . klDivergence_ ); // Visualize the results embedding . forEach (( point , i ) => { console . log ( `Point ${ i } : [ ${ point [ 0 ]. toFixed ( 3 ) } , ${ point [ 1 ]. toFixed ( 3 ) } ]` ); });
Notes
TSNE is primarily used for visualization, not for general dimensionality reduction
Results can vary between runs; use randomState for reproducibility
Perplexity should be smaller than the number of samples
The algorithm preserves local structure but may distort global structure