The main chapters of this book teach by guided invention: you derive each algorithm yourself, one small step at a time, before the concept is named. These 47 exercises are a different mode — drill practice. Each notebook gives you one concrete formula or algorithm and asks you to implement it in Python from scratch. There is no arc or narrative; every notebook stands alone.
When to use these. Work through the five groups in order — Expressions, If/Else, Loops, Functions, Recursion — as you finish the corresponding LBI chapter, or as warm-ups before the ML chapters. The ML themes (gradient descent, entropy, neural networks) preview concepts you will invent in full later; if a problem feels too hard, come back after the relevant chapter.
How each notebook works. Read the problem statement.
Fill in the starter cell wherever you see # TODO. Run the
test cell at the bottom — if it prints All tests passed!,
you’re done.
Best done after: Chapter 3 (Expressions & Functions)
Each problem is a single Python expression. No loops, no conditionals
— just arithmetic, math.exp, and one or two variables.
| # | Difficulty | Concept | Open in Colab |
|---|---|---|---|
| 1 | Easy | Mean of three numbers | |
| 2 | Easy | Z-score for one value | |
| 3 | Easy | Min-max scale one value | |
| 4 | Medium | Euclidean distance (2-D) | |
| 5 | Medium | Probability of two independent events | |
| 6 | Medium | Squared error for one prediction | |
| 7 | Medium | Sigmoid activation function | |
| 8 | Hard | One term of a covariance sum | |
| 9 | Hard | One gradient descent update step | |
Best done after: Chapter 4 (If/Else)
Each problem requires a conditional choice — "A" vs
"B", "pure" vs "mixed",
"train" vs "val" vs "test".
Threshold-based decisions are everywhere in ML.
| # | Difficulty | Concept | Open in Colab |
|---|---|---|---|
| 1 | Easy | Flag a single outlier by z-score | |
| 2 | Easy | Classify a probability as likely or unlikely | |
| 3 | Easy | Turn a sigmoid output into a class label | |
| 4 | Medium | Classify one prediction as TP / FP / TN / FN | |
| 5 | Medium | Assign a point to its nearest centroid | |
| 6 | Medium | Classify correlation strength and direction | |
| 7 | Medium | Min-max normalize without dividing by zero | |
| 8 | Hard | Assign a sample to train / val / test by position | |
| 9 | Hard | Decide whether a tree node is pure | |
Best done after: Chapter 8 (Loops & Arrays)
Each problem asks you to accumulate a result by looping over a list. The themes run from basic statistics (mean, std dev) up to gradient descent and entropy — you will invent the full versions of those ideas in Chapters 15 and 21.
| # | Difficulty | Concept | Open in Colab |
|---|---|---|---|
| 1 | Easy | Compute the mean with a loop | |
| 2 | Easy | Sum of squared deviations | |
| 3 | Medium | Compute the standard deviation | |
| 4 | Medium | Find all outliers in a dataset | |
| 5 | Medium | Min-max normalize an entire list | |
| 6 | Medium | Distances from a query point to a training set | |
| 7 | Hard | Build a confusion matrix over a dataset | |
| 8 | Hard | Compute covariance between two lists | |
| 9 | Hard | Run several iterations of gradient descent | |
| 10 | Hard | Compute the entropy of a list of labels | |
Best done after: Chapter 3 (Expressions & Functions) and Chapter 8 (Loops & Arrays)
These problems ask you to compose smaller functions into larger ones — the same building-block pattern the book uses throughout. The capstone notebooks (regression pipeline, entropy + information gain, sigmoid + perceptron) are mini-versions of full LBI chapters.
| # | Difficulty | Concept | Open in Colab |
|---|---|---|---|
| 1 | Easy | mean() function |
|
| 2 | Easy | variance() using mean() |
|
| 3 | Easy | std_dev() using variance() |
|
| 4 | Medium | find_outliers() from stats functions |
|
| 5 | Medium | Simple k-nearest-neighbors classifier | |
| 6 | Medium | precision(), recall(), and
f1_score() |
|
| 7 | Medium | correlation() from covariance and std dev |
|
| 8 | Hard | Linear regression pipeline (predict, mse,
gradient_descent_fit) |
|
| 9 | Hard | entropy() and information_gain() |
|
| 10 | Hard | sigmoid() and perceptron() |
|
Best done after: Chapter 5 (Recursion)
Each problem has a natural recursive structure: a base case and a case that reduces to a smaller version of itself. The ML themes (learning-rate decay, tree depth) appear in the Gradient Descent and Decision Tree chapters.
| # | Difficulty | Concept | Open in Colab |
|---|---|---|---|
| 1 | Easy | Recursive sum of a list | |
| 2 | Easy | Recursive product of a list (geometric mean) | |
| 3 | Medium | Recursive sum of squared deviations | |
| 4 | Medium | Recursive factorial (for counting outcomes) | |
| 5 | Medium | Binomial coefficient via Pascal’s rule | |
| 6 | Medium | Binary search over sorted distances | |
| 7 | Hard | Recursive merge sort (to find the median) | |
| 8 | Hard | Recursive learning-rate decay schedule | |
| 9 | Hard | Recursive depth of a decision tree | |