Can UMAP be used for clustering?

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

Does UMAP require data normalization?

Yes, it is highly recommended to standardize features to comparable scales (e.g., using StandardScaler) before applying UMAP to ensure equal weighting across all dimensions.

Why should I use UMAP instead of t-SNE?

UMAP is generally significantly faster than t-SNE, scales better to large datasets, and preserves more of the global structure of the data, whereas t-SNE focuses almost exclusively on local structure.

How do I choose the best n_neighbors value?

Start with the default of 15. Use lower values (2-5) to capture fine local detail or higher values (50-200) to prioritize the broad topological structure of the entire dataset.

UMAP Dimensionality Reduction

Name: UMAP Dimensionality Reduction
Author: youyinnn

byyouyinnn

0•

データサイエンスとML

Implements fast non-linear dimensionality reduction for high-dimensional data visualization and clustering preprocessing.

This skill provides comprehensive guidance on using UMAP-Learn to transform complex high-dimensional datasets into lower-dimensional representations while preserving both local and global topological structures. It offers ready-to-use patterns for 2D/3D visualization, optimized preprocessing workflows for clustering algorithms like HDBSCAN, and advanced techniques for supervised or parametric dimensionality reduction using neural networks. Whether you are performing exploratory data analysis or building machine learning pipelines, this skill ensures efficient and scalable embedding generation.

主な機能

01Seamless integration with scikit-learn pipelines and HDBSCAN clustering

02Parametric UMAP implementation using deep learning for new data transformation

030 GitHub stars

04Fast non-linear dimensionality reduction preserving local and global structure

05Extensive distance metric support including Euclidean, Cosine, and Manhattan

06Support for supervised and semi-supervised embedding generation

ユースケース

01Preprocessing datasets to overcome the curse of dimensionality for clustering

02Visualizing high-dimensional clusters and data relationships in 2D or 3D

03Supervised feature engineering for downstream machine learning classifiers

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add youyinnn/skills-collection umap-learn

For use in Claude.ai and ChatGPT

主な機能

01Seamless integration with scikit-learn pipelines and HDBSCAN clustering

02Parametric UMAP implementation using deep learning for new data transformation

030 GitHub stars

04Fast non-linear dimensionality reduction preserving local and global structure

05Extensive distance metric support including Euclidean, Cosine, and Manhattan

06Support for supervised and semi-supervised embedding generation

ユースケース

01Preprocessing datasets to overcome the curse of dimensionality for clustering

02Visualizing high-dimensional clusters and data relationships in 2D or 3D

03Supervised feature engineering for downstream machine learning classifiers