Learning path
Next moduleModule 8: Dimensionality reduction
Module 7
Unsupervised learning
Clustering and anomaly detection without labels, and how to tell a real group from an artifact.
- 1. K-means clusteringGroup unlabeled data by alternating assignment and update steps, compute an iteration by hand, and recognize when K-means' assumptions fail.35 min
- 2. DBSCAN, hierarchical clustering, and Gaussian mixturesChoose a clustering method by the shape and density of your clusters, run DBSCAN by hand, read a dendrogram, and use soft memberships and anomaly scores.30 min
With no labels, the goal is structure: groups, densities, and outliers. Every method here makes assumptions about what a cluster looks like. Learn the assumptions and you will know when to trust the output.
| # | Lesson | You will be able to |
|---|---|---|
| 1 | K-means | run assignment and update steps by hand and choose K |
| 2 | DBSCAN, hierarchical clustering, and Gaussian mixtures | pick a clustering method by cluster shape and density, and use soft memberships |
Interactive lab: K-means
Key distinctionA discovered cluster is a hypothesis, not a verified real-world category.