---
id: umap-learn
version: "0.5.12"
license: BSD
license_treatment: permissive
maintenance: active
---

# umap-learn — Uniform Manifold Approximation and Projection

License: permissive · Maintenance: active · Downloads: 7.5M/mo

## What it is and what it does

UMAP is a dimension reduction algorithm that embeds high-dimensional data into lower-dimensional space while preserving manifold structure. It models data as lying on a Riemannian manifold and uses fuzzy topological structure to find embeddings that balance local neighborhood relationships with global structure. The package follows a familiar transformer API, integrating naturally into existing ML pipelines.

The algorithm depends on numpy, scipy, scikit-learn, numba (for performance), pynndescent (for nearest-neighbor search), and tqdm (for progress reporting). It supports sparse matrices, custom distance metrics, and includes densMAP for density-preserving embeddings. You can use it for exploratory visualization, as a preprocessing step before clustering, or as a general-purpose non-linear dimensionality reduction technique.

Use it for:

- Visualize high-dimensional datasets (e.g., image pixels, gene expression) in 2D or 3D for exploratory data analysis
- Preprocess data before clustering to improve cluster quality
- Reduce dimensionality as a feature engineering step before training downstream models
- Preserve local density information in embeddings using densMAP for single-cell transcriptomics or similar applications
- Handle sparse, very high-dimensional data where alternatives become computationally prohibitive

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

UMAP reduces high-dimensional data to lower dimensions for visualization and general non-linear dimensionality reduction, using a manifold-learning approach that preserves both local and global structure.

Yes. UMAP is actively maintained, has no known vulnerabilities, low install friction, and a permissive BSD license. It is widely used in scientific and ML workflows. Install it if you need dimension reduction with better global structure preservation, or as a general-purpose manifold learning tool for exploratory analysis and preprocessing.

## Install

pip install umap-learn
uv add umap-learn
poetry add umap-learn

## Installing umap-learn

Before you install: Low install friction with a pure-Python wheel distribution. Requires numba for JIT compilation and dependencies including numpy and scipy, all of which are well-maintained. Active maintenance with a recent release 128 days ago.

License in practice: BSD permissive license allows use in commercial and proprietary projects with minimal restrictions; you may use, modify, and distribute with proper attribution.

Quickstart:

pip install umap-learn

import umap

embedding = umap.UMAP().fit_transform(data)

Requires Python 3.9 or later; numba JIT compilation may take time on first run but subsequent calls are fast.

Verify before relying:

- Whether the package supports GPU acceleration or if computation is CPU-only via numba
- Performance characteristics on datasets with over a million dimensions (mentioned in description but not quantified)
- Whether densMAP mode requires additional configuration or dependencies beyond the base install

## Package facts

- License: BSD (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 7.5M/month (top 5,000 on PyPI)
- Known vulnerabilities: none known

## Tags

dimension reduction visualization, manifold learning, high-dimensional data embedding, t-SNE alternative, non-linear dimensionality reduction, data visualization projection, topological data analysis, dimensionality-reduction, manifold-learning, visualization

[View on SkillFed](https://skillfed.io/packages/umap-learn) · [View on PyPI](https://pypi.org/project/umap-learn/)