Module 26

Learning without labels: clustering and dimensionality reduction

Advanced module (after graduating): k-means, PCA, outliers

When you finish this module you can say:

I can find clusters in unlabeled data with k-means, reduce its dimensions with PCA and separate outlying points; I know I am measuring usefulness, not correctness.

By the end of this module you will be able to cluster an unlabeled table, reduce its dimensions and find outlying rows, and explain why none of these results is a single right answer.

Lessons

6 lessons · ≈ 66 min

1. What is learned without labels10 min
2. k-means: assign, move, repeat11 min
3. How many clusters? Elbow, silhouette, scale11 min
4. Dimensionality reduction: PCA11 min
5. Finding the outlier11 min
6. Project: cluster your notes12 min · project

Practice exercises

Short writing exercises, independent of the lessons: Go to the exercises

1. The point nearest to the center
2. How many components are enough?
3. The positions of the outlying rows