Hierarchical and density-based clustering
Two ways to cluster without centroids. Merge the closest groups step by step and keep the whole family tree, or grow clusters through dense regions and call everything else noise. Linkage criteria, dendrograms, DBSCAN and HDBSCAN.
50 min
After this lesson you can
- Run agglomerative clustering by hand and read a dendrogram, including where to cut it
- Predict how single, complete, average and Ward linkage behave, and derive Ward's merge cost
- Explain why single linkage is a minimum spanning tree, and why it chains
- Define DBSCAN's core, border and noise points and choose ε with the k-distance plot
- Say what DBSCAN can't do (clusters of different densities) and how HDBSCAN addresses it
In this lesson
- Merging, one pair at a time
- Density: clusters are crowded places
- Clusters of different densities: HDBSCAN
This lesson is part of Descent Pro
Descent Pro unlocks every lesson on the route, from base camp to the summit. Every plan starts with a free trial.