Class Tsne
java.lang.Object
ai.beestream.jitnn.ml.Tsne
Exact t-SNE (van der Maaten and Hinton, 2008) on the device, after scikit-learn's TSNE with
method="exact": Gaussian affinities calibrated per point to the perplexity by bisection on the
precision, symmetrised; a Student-t kernel in the embedding; gradient descent with momentum
(0.5, then 0.8), early exaggeration (12 for 250 iterations), per-parameter gains (+0.2 / x0.8,
at least 0.01) and the learning rate max(n / 12 / 4, 50). The embedding starts from PCA, scaled
so the first coordinate has standard deviation 1e-4. O(n^2) memory and time per iteration.
-
Constructor Summary
Constructors -
Method Summary
Modifier and TypeMethodDescriptionThe embedding [n, components] of samples x [n, features], starting from their PCA projection.
-
Constructor Details
-
Tsne
public Tsne(int components, float perplexity, int iterations) t-SNE intocomponentsdimensions with the given perplexity and number of iterations. -
Tsne
public Tsne()t-SNE with scikit-learn's defaults: 2 dimensions, perplexity 30, 1000 iterations.
-
-
Method Details
-
fitTransform
-