STA5077Z Exam Book
Unsupervised Learning — closed-book study notes
How to use this book
Study the Pass and the Recall of each chapter. Open Reference only when a question beats you. Nothing here enters the exam room.
Exam constraints
| Fact | Detail |
|---|---|
| When | Wednesday 11 November, 09:00–14:00 (5 hours) |
| Notes | No notes. Same conditions as the test. |
| Split | 50 marks dimension reduction (Miguel), 50 marks clustering (Andrew) |
| Clustering written | 60% of that half ≈ 30 marks; about two-thirds hierarchical |
| Clustering coding | 40% ≈ 20 marks, split equally k-means / hierarchical |
| DR theory | 25–30 marks; question bank is the guide |
| DR coding | 20–25 marks; main function, extract, interpret |
| Clustering packages | base R and cluster only |
| Templates | Scaffolded .qmd / .Rmd; plain-text answers; code in chunks |
| Render failure | Set eval: false or hand in source — no mark penalty |
| SOM coding | Not examined (theory only) |
| Simulation / loops | Not asked in coding |
Not examinable
- K-medoids and PAM (Andrew: Lecture 1–2, slides 24–32)
- DBSCAN
- Market-basket analysis
Last year’s paper is used only where it still matches this year’s topic list.
Study order (about 20 hours)
- Shared clustering ideas — 1h
- K-means — 2.5h
- Hierarchical — 3.5h
- PCA — 3h
- MDS — 2h
- ISOMAP — 2h
- SOM — 3h
- Autoencoders — 2h
- Comparisons + blank rehearsal — 1h
Chapter pattern
Each idea in a Pass appears as: situation → mechanism → formula or command → one worked number. A formula without a reason is incomplete for this exam.