umap

Why we included this project

UMAP reduces high-dimensional data to two or three dimensions while keeping the structure that matters. It plugs into scikit-learn pipelines as a drop-in replacement for t-SNE, so anyone comfortable with sklearn's transformer API can start embedding with a single fit_transform call. The same embedding works as a preprocessing step before clustering or as a feature for downstream models, and the densMAP variant preserves local density information for single-cell and other density-sensitive work. The n_neighbors and min_dist parameters give direct control over how much global versus local structure the embedding keeps. Documentation is solid, and.