smallvis

Installing

install.packages("devtools")devtools::install_github("jlmelville/smallvis/smallvis")library(smallvis)

Prerequisites

• Thevizier package to plot the coordinates during optimization. It’s not on CRAN, and therefore requires a fairly new version ofdevtools (1.9 or greater) to install this as a dependency from github.
• Similarly, themize package (which is on CRAN but not in a sufficiently advanced state currently).
• Optionally,RSpectra package, which is used only if you want to initialize from a spectral method (Y_init = "laplacian" orY_init = "normlaplacian"). If not present, then the standard R functioneigen is used, but it’s alot slower.

Explorations ofsmallvis

I have written some longer form documents that demonstrate the output ofsmallvis and its various options, with an emphasis on t-SNE.

• Thedatasets used in my ruminations.
• Sometheory showing a comparison of cost functions and gradients.
• Some material on variousspectral methods, which justifies the use of Laplacian Eigenmaps for initialization (a bit).
• A look at the effect of thescaling andPCA and whitening preprocessing options on t-SNE.
• Howperplexity affects t-SNE.
• Usingintrinsic dimensionality toautomatically select perplexity. Also,non-global perplexities.
• Distance-preserving methods (geodesic as well as Euclidean).
• A comparison of various flavours ofStochastic Neighbor Embedding (t-distributed and otherwise).
• A comparison ofNeRV and JSE and a look at symmetric JSE (SJSE).
• Heavy-Tailed SSNE (HSSNE).
• Optimizing the Heavy-Tail parameter in HSSNE.
• Testing some differentinitialization methods for t-SNE. Related: does using anaverage distance matrix based on the result of multiple random initializations do better than just picking the result with the lowest error? (It doesn’t.)
• A look at an exact gradient version ofLargeVis.
• Some initial results usingUMAP and related methods. Also, looking at the effect of borrowing some features from UMAP to seethe effect on t-SNE (sadly, not much).
• t-SNE normalization.
• An evaluation oflate exaggeration in t-SNE, as suggested in theFIt-SNE paper.
• Elastic Embedding and at-Distributed variant.
• Optimizaton: L-BFGS (and qSNE).
• Optimizaton: Spectral Direction.
• Optimizaton: Conjugate Gradient.
• Optimizaton: SGD methods.
• Perplexity and three clusters from fromHow to Use t-SNE Effectively.
• How hard is it to unroll theSwiss Roll with SNE?
• Using input bandwidths inSNE output functions.
• Multiscale perplexities as used in multiscale t-SNE.
• Theopt-SNE settings applied to smaller sized datasets.
• Global SNE (g-SNE).
• t-SNE using f-Divergences.
• ABSNE.
• Related: abibliography of interesting (to me) t-SNE and variants.