Tutorials · Chapter D (4/4) · ~10 min

Clustering

Try it → see it → read → next

Group similar examples without labels, then explain what the groups do and do not mean.

Try yourself

Playground

Cluster playground

Cluster the points, then assign meaning. Don’t mistake a grouping for ground truth.

Recap

What you just did

ClusterPlayground grouped unlabeled points, then tempted you with a permanent label. Clusters are hypotheses to interpret carefully, not ground truth.

Teach

How it works

A k-means-style loop is surprisingly small:

points = [(1, 1), (2, 1), (8, 8), (9, 8)]
centers = [(1, 1), (9, 8)]

# Repeat:
# 1) assign each point to its nearest center
# 2) move each center to the mean of its assigned points
# 3) stop when assignments barely change

Features place examples on a map
Distance defines what “similar” means
Assignment connects each point to a cluster
Center update makes groups settle

Mental model: clustering arranges a messy drawer into piles; you still have to decide what each pile means.

Use it

When you'd use this

Finding rough customer or document segments
Exploring an unlabeled dataset before supervised training
Grouping embeddings by semantic similarity

Watch out

Clusters are not natural facts or fair human categories. Changing feature scales can completely change the groups. Never turn exploratory clusters into sensitive labels without evidence, domain review, and bias checks.

Try next

Try this next

Imagine multiplying one feature by 100. Predict which direction would dominate distance, then normalize it and compare.