title: “Reading companion”
Reading companion
Reading companionThese chapters follow the same story as the lecture decks, but at a slower pace. Use them before class, after class, or when you miss a session and need the ideas explained in plain language.
We keep two example tracks across the week: simulated genes (where we know the ground truth) and Palmer Penguins (real measurements that are messier and more realistic). For context on every dataset, use the data catalog.
If you want the overall narrative first, read the curriculum outline. If you want the long coding versions, jump to the student notebooks.
| Chapter | Teaching day | What you will understand | Open |
|---|---|---|---|
| Chapter 1 — Learning from measurements | Day 1 (Mon) | Why train/validation/test splits matter and how to avoid leakage | Open |
| Chapter 2 — When coefficients need restraint | Day 1 (Mon) | Ridge, lasso, and elastic net in plain language | Open |
| Chapter 3 — Rules and trees | Day 2 (Tue) | How tree models split data and why “two cultures” helps | Open |
| Chapter 4 — A reproducible modeling workflow | Day 2 (Tue) | How recipe, workflow, last_fit, and tuning fit together |
Open |
| Chapter 5 — Stronger learners, same discipline | Day 4 (Thu) | How random forests, boosting, and MLPs fit into the same pipeline | Open |
| Chapter 6 — Scores that match the question | Day 4 (Thu) | How to choose metrics under imbalance and read SHAP cautiously | Open |
| Chapter 7 — Choosing what to optimize | Day 4 / lab | Why different tuning metrics pick different “best” settings | Open |
| Chapter 8 — Comparing learners fairly | Day 4 / lab | How to compare model families fairly with fixed folds and recipe | Open |
Chapter 1 — Learning from measurements
This chapter sets up the full course: prediction as a disciplined process, not just fitting a line. You will see the same logic on simulated genes and penguins.
Chapter 2 — When coefficients need restraint
This chapter explains why regularization helps when predictors are correlated or numerous, and how ridge, lasso, and elastic net differ in practice.
Chapter 3 — Rules and trees
This chapter introduces decision trees as nested if-then rules and places them in the broader “two cultures” view: assumptions-first versus prediction-first.
Chapter 4 — A reproducible modeling workflow
This chapter turns modeling into a repeatable workflow: one honest train/test example first, then the same tuned tree pipeline used in Day 2 slides.
Chapter 5 — Stronger learners, same discipline
This chapter explains what changes and what stays fixed when you move from single trees to random forests, boosting, and small neural nets.
Chapter 6 — Scores that match the question
This chapter focuses on decision-aware evaluation: choosing metrics for the question at hand, handling class imbalance, and reading SHAP with causal caution.
Chapter 7 — Choosing what to optimize
This extension chapter shows how metric choice changes tuning outcomes, even when you keep data, folds, and model family fixed.
Chapter 8 — Comparing learners fairly
This extension chapter compares three stronger learners fairly by holding preprocessing and folds fixed and changing only the model engine.