What is the main advantage of UMAP over t-SNE?

UMAP is significantly faster and more scalable than t-SNE while typically better preserving the global structure of the data, making it more suitable for large-scale datasets.

Does this skill support transforming new, unseen data?

Yes, it provides implementation patterns for the transform() method in standard UMAP and detailed guides for using Parametric UMAP for neural-network-based projections.

Is supervised dimensionality reduction supported?

Yes, this skill includes patterns for passing label information to the fit method, allowing the embedding to be guided by class information for better separation.

Can UMAP be used for clustering high-dimensional data?

Yes, UMAP is an ideal preprocessing step for density-based clustering like HDBSCAN. For best results, set min_dist to 0.0 and increase n_components to between 5 and 10.

How do I choose the right value for n_neighbors?

Use lower values (2-5) to focus on fine local detail and higher values (50-200) to prioritize the broad topological structure; 15 is generally the best starting point.

UMAP Dimensionality Reduction

Name: UMAP Dimensionality Reduction
Author: plurigrid

byplurigrid

•

데이터 과학 및 ML

Performs fast, scalable non-linear dimensionality reduction for high-dimensional data visualization, clustering preprocessing, and feature engineering.

This skill integrates the UMAP (Uniform Manifold Approximation and Projection) library into Claude's workflow, enabling sophisticated analysis of complex datasets through manifold learning. It provides comprehensive patterns for preserving both local and global data structures, tuning critical parameters like neighborhood size and minimum distance, and implementing advanced supervised or parametric embeddings. Whether you are creating 2D/3D visualizations, preparing data for density-based clustering with HDBSCAN, or building machine learning pipelines, this skill provides the domain-specific guidance needed to handle high-dimensional data effectively.

주요 기능

01Parametric UMAP integration for deep learning and efficient new data projection

028 GitHub stars

03Optimized preprocessing for density-based clustering algorithms like HDBSCAN

04Comprehensive parameter tuning for local vs. global structure preservation

05Non-linear dimensionality reduction for 2D and 3D data visualization

06Supervised and semi-supervised embedding support using label information

사용 사례

01Preprocessing noisy data to identify distinct clusters in non-linear manifolds

02Visualizing complex high-dimensional datasets such as genomic data or word embeddings

03Reducing feature space dimensionality to improve the performance of downstream classifiers

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add plurigrid/asi umap-learn

For use in Claude.ai and ChatGPT

Download Skill