smallvis is an R package for experimenting with neighborhood embedding methods, including t-SNE, LargeVis, and UMAP cost functions. You can control input scaling, nearest neighbors, initialization, and optimization to compare methods and try out ideas.
Install from GitHub with pak::pak("jlmelville/smallvis/smallvis"). The R package lives in the smallvis subdirectory of the repository.
Start with Using smallvis to make and plot an embedding. Continue with the guides to inputs and initialization, optimization and repeated runs, or Barnes-Hut and native acceleration.
The article collection brings together the theory and experiments behind the package, including spectral methods, perplexity, initialization, optimization, and comparisons with other embedding methods. Dated explorations retain their original results and examples. The function reference documents the current arguments and return values.