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

UMAP is generally faster, scales more efficiently to large datasets, and preserves a better balance of local and global topological structures.

How should I tune UMAP for clustering purposes?

For clustering (like HDBSCAN), it is recommended to increase n_neighbors to around 30, set min_dist to 0.0, and use a higher n_components value (5-10) to preserve density.

Does this skill support supervised learning?

Yes, UMAP can incorporate target labels during the fitting process to create embeddings that emphasize class separation while maintaining internal structure.

Can UMAP handle new data points after training?

Yes, UMAP provides a transform() method that allows you to project new data into the existing embedding space without retraining the entire model.

UMAP-Learn Dimensionality Reduction

Name: UMAP-Learn Dimensionality Reduction
Author: BbgnsurfTech

byBbgnsurfTech

•

データサイエンスとML

Performs fast, nonlinear dimensionality reduction to visualize high-dimensional data and optimize machine learning clustering workflows.

The UMAP-Learn skill integrates the Uniform Manifold Approximation and Projection algorithm into Claude's workflow, providing a powerful tool for analyzing high-dimensional datasets. It excels at preserving both local and global structures within data, making it superior to t-SNE for many visualization and preprocessing tasks. This skill enables Claude to guide you through complex tasks such as 2D/3D data visualization, supervised metric learning, and preprocessing for density-based clustering (HDBSCAN), while offering specific guidance on parameter tuning for n_neighbors, min_dist, and metric selection.

主な機能

01Preserves local and global data structure better than t-SNE

02Efficient transformation of new, unseen data into learned spaces

03Supports supervised and semi-supervised dimensionality reduction

04Optimized preprocessing for density-based clustering like HDBSCAN

053 GitHub stars

06Parametric UMAP support for deep learning integration

ユースケース

01Visualizing high-dimensional embeddings from NLP or genomic data

02Clustering complex datasets by reducing noise and dimensions

03Feature engineering to reduce dimensionality before training classifiers

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add bbgnsurftech/claude-skills-collection umap-learn

For use in Claude.ai and ChatGPT

Download Skill