Umap Learn

UMAP is a dimensionality reduction algorithm that's faster than t-SNE and actually scales beyond 2D visualization. You'd use this when you need to compress high-dimensional data for plotting, preprocessing before HDBSCAN clustering, or feature engineering in ML pipelines. The key insight here is that visualization settings (n_neighbors=15, min_dist=0.1, 2D) are different from clustering settings (n_neighbors=30, min_dist=0.0, 5-10D). It supports supervised and semi-supervised modes by passing labels during fitting, which is surprisingly useful for metric learning. Always standardize your input data first. The documentation does a solid job explaining the four main parameters and why you'd tune them differently depending on whether you care about local detail or global structure.

skills.sh installs: 2.1K
GitHub stars: 56.6K
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add k-dense-ai/scientific-agent-skills --skill umap-learn --agent claude-code

Installs into .claude/skills of the current project.

Files

UMAP-Learn

Overview

UMAP (Uniform Manifold Approximation and Projection) is a dimensionality reduction technique for visualization and general non-linear dimensionality reduction. Apply this skill for fast, scalable embeddings that approximate neighborhood structure, supervised learning, and clustering preprocessing.

Before interpreting an embedding, check finite coordinates, neighborhood retention, sensitivity to seeds/parameters, and domain evidence in the original space. UMAP axes, island areas, and gaps between clusters have no calibrated physical or probabilistic meaning. Keep predictive test data outside preprocessing and embedding fits.

Quick Start

Installation

Verified on 2026-10-01: umap-learn 0.5.12 is the current stable release (April 2026). Requires Python 3.9+ and depends on scikit-learn>=1.6, numba, pynndescent, numpy, and scipy. Pin to a verified release: