Commute learning plan
Use flexible study blocks, not a deadline-dependent calendar. A block is about 25–35 minutes. If your session is shorter, split a block and pick up where you left off.
What to do in each block
- Overview, 5 minutes: get the lecture’s question and main concepts.
- Explain, 3 minutes: look away and describe the main distinctions aloud or on paper. Note what remains unclear.
- Deep dive, 12–18 minutes: read the relevant detailed sections; follow worked examples with your own arithmetic or trace code on paper. Use a visual page where available.
- Practice, 5–8 minutes: attempt one original question before revealing the answer/rationale. Record the mistake or uncertainty in your notes.
- Finish, 2 minutes: write a few points you can now explain and one section to revisit.
The second question can be attempted at the start of the next journey, before rereading. This separates understanding an explanation from recalling it later.
Start with the foundations: lectures 1–4
| Block | Focus | Evidence of learning |
|---|---|---|
| 1 | Lecture 1 | Explain the three disciplines, data-to-wisdom ladder and lifecycle |
| 2 | Lecture 2 | Trace vectors, missing values, indexing and functions; reproduce the small calculation yourself |
| 3 | Lecture 3 | Explain API/scraping, ETL and warehouse/mart/lake with a sport example |
| 4 | Lecture 4 | Trace a grouped summary/join, distinguish scales, and interpret a boxplot/ggplot mapping |
Then the methods: lectures 5–9
| Block | Focus | Evidence of learning |
|---|---|---|
| 5 | Lecture 5 overview and feature operations | Explain aggregation, binning, encoding and combining with different examples |
| 6 | Lecture 5 errors and regularization | Calculate MAE/MSE; explain λ, ridge/LASSO and unseen-data evaluation; finish its questions |
| 7 | Lecture 6 classification, fit and validation | Identify positive class; calculate precision/recall; diagnose a train/validation gap |
| 8 | Lecture 6 clustering and PCA | Step through k-means; distinguish elbow/silhouette and variance from accuracy; finish its questions |
| 9 | Lecture 7 scaling and missingness | Calculate z-score/min–max; explain why imputation and deletion need judgment |
| 10 | Lecture 7 risk and networks | Calculate risk, odds, RR/OR and communicate the absolute difference; finish its questions |
| 11 | Lecture 8 | Draw the exposure-model → linkage → health-model pipeline; interpret HR with limits |
| 12 | Lecture 9 | Explain validation, context/reference data, noise versus missing hard efforts, and selection |
If a method relies on a basic you have not mastered yet, open that lecture's overview first. For R practice, a script and an installed R environment are enough; the notes also support tracing code and writing outputs without running it.
Make it stick across journeys
- On the next journey, recall the previous lecture before opening it, then attempt its remaining question.
- After roughly three days, revisit the distinctions you confused and redo the relevant calculations without the worked answer.
- After roughly a week, mix topics: modelling/evaluation from 5–6, preprocessing from 4/7, and data limitations from 3/8/9.
- Mark each lecture for yourself: “can explain”, “needs a prompt”, or “needs another example”. Revisit the latter two first.
Do not use reading speed as the success measure. Being able to explain a method choice, trace a short script or justify a conclusion is more useful.
Use the real example exams later
Once you have worked through the practice questions and the question bank, use the official example exams on Canvas for an exam-style session. Cover answer markings/avoid the key, attempt independently, then compare the reasoning and identify the lecture section behind each mistake. Older guest examples can differ from this year’s slides; they inform style, not a promise of which questions will appear.