Week 15: Finding Patterns Without Labels

Supervised learning begins with target answers. Unsupervised learning asks what structure can be found when those labels are absent.

This week uses clustering to expose an important limitation: a group produced by an algorithm is not automatically a natural category in the world.

You will trace the assign-and-update loop in k-means, examine how distance controls similarity, and then change feature scaling and outliers in the Clustering Playground.

The goal is not to make the clusters look convincing. The goal is to defend what the grouping is useful for—and to refuse names or conclusions the data does not support.