Clustering Pipelines

A frequent task in data science is identifying clusters in high-dimensional space and then visualizing the learned structural manifolds. This showcase allows you to run multiple clustering algorithms—such as K-Means, OPTICS, CURE, and Hierarchical Clustering—on various datasets. It computes the assignments intrinsically and then uses UMAP for the 2D visual projection.

Pipeline Settings

DatasetIrisWinePenguinsBlobs

AlgorithmK-MeansK-MedoidsX-MeansCUREMean ShiftOPTICSHierarchical Clustering

Clusters (K): 3

This example runs KMeans on the original high-dimensional data, assigning cluster labels locally, and then uses UMAP to visualize the manifold in 2D.

How-to (Code)

Combining clustering and dimensionality reduction is a common pattern for exploratory data analysis. You can use any of DruidJS's clustering algorithms for the assignments and a technique like UMAP for visualization.

import * as druid from "@saehrimnir/druidjs";

const data = [
  /* ... multi-dimensional float array ... */
];

// 1. Run a clustering algorithm (e.g., Hierarchical Clustering)
const hc = new druid.HierarchicalClustering(data, { linkage: "complete" });
const clusters = hc.get_cluster_list(5.0); // Cut the dendrogram tree at distance 5.0

// Or use OPTICS
// const optics = new druid.OPTICS(data, { epsilon: 1.0, min_points: 4 });
// const clusters = optics.get_cluster_list();

// 2. Run UMAP for visualization
const umap = new druid.UMAP(data);
const projection = umap.transform();

// Combine `clusters` and `projection` for plotting...