openTSNE
An extensible, parallel Python implementation of t-SNE, a dimensionality-reduction algorithm, with support for adding new points to an existing embedding and scaling to millions of data points.
openTSNE is a modular Python implementation of t-distributed stochastic neighbor embedding (t-SNE), a popular algorithm for visualizing high-dimensional datasets in two or three dimensions. It incorporates recent improvements to the algorithm, including adding new data points to an existing embedding, substantial speed improvements that let it scale to millions of points, and techniques for better preserving the global alignment of the resulting visualization.
It installs via conda-forge or pip, and building from source requires a C/C++ compiler; installing FFTW3 before building improves performance on large datasets by accelerating the Fast Fourier Transform the algorithm relies on, falling back to numpy's FFT implementation otherwise.
Documentation includes a user guide, tutorial, basic and advanced examples, and speed benchmarks.