UMAP-JS Dimension Reduction

UMAP-JS is a JavaScript implementation of the Uniform Manifold Approximation and Projection (UMAP) algorithm, a popular dimension reduction technique for both visualization and general non-linear dimension reduction. Currently stable at version 1.4.0, the library offers synchronous, asynchronous, and step-by-step fitting methods, allowing for flexible integration into various application contexts. Key differentiators from the original Python implementation include the use of a random embedding to seed the optimization step, rather than a spectral embedding, which yields comparable results for smaller datasets but avoids complex eigenvalue/eigenvector computations in JavaScript. The library also lacks specialized functionality for angular distances or sparse data representations. Releases appear to follow an irregular schedule, with the last publish about two years ago, but the project is maintained by PAIR-code, indicating ongoing support.