umap-learn

UMAP reduces high-dimensional data to 2D or 3D for visualization while preserving both local and global structure. This Skill performs fast nonlinear manifold learning, supporting 15+ distance metrics for accurate data representation.

Yes, using UMAP for clustering preprocessing prepares data for more effective clustering algorithms like HDBSCAN. It compresses features into a lower-dimensional space, making subsequent clustering more efficient and accurate.

UMAP integrates seamlessly into scikit-learn pipelines for machine learning tasks. It supports feature engineering by creating lower-dimensional features to train models within standard scikit-learn workflows.

To visualize thousands of gene expression features, use nonlinear manifold learning to reduce them to 2 components. This Skill compresses the data so you can plot it and visually identify distinct cell populations.

UMAP supports 15+ distance metrics for manifold learning. This compatibility allows it to handle various data types and structures during dimensionality reduction and feature engineering.

Available UMAP variants include Parametric UMAP, DensMAP, and AlignedUMAP. These options extend standard nonlinear manifold learning to support specialized visualization and feature engineering needs.