umap-learn

Manifold learning reduces high-dimensional data to meaningful low-dimensional embeddings for visualization. UMAP provides fast, meaningful embeddings applicable across data analytics workflows and feature engineering.

UMAP supports supervised and semi-supervised embedding capabilities alongside standard dimensionality reduction. This allows you to incorporate labels into the embedding process for more targeted feature engineering.

UMAP supports transforming new data and integrates into machine-learning pipelines. You can fit a model on existing data and apply the transform method to project new samples into the same low-dimensional space.

Standard preprocessing steps like scaling are required before applying dimensionality reduction. UMAP requires Python and the umap-learn library to execute these scaling and transform operations effectively.

Parametric UMAP supports neural networks for dimensionality reduction. This requires optional dependencies beyond the standard umap-learn installation to enable neural network-based embeddings.

UMAP provides fast, meaningful embeddings specifically designed for clustering preprocessing and visualization. It distinguishes itself from other dimensionality reduction tools by preserving both local and global manifold structure effectively.