Definition
Ready to show mastery of supervised learning?
This assessment covers Supervised Machine Learning in Depth: the workflow and leakage, preprocessing and pipelines, linear and logistic regression with their derivations, k-NN, Naive Bayes, SVMs, trees and ensembles, evaluation metrics and curves, imbalance, cross-validation, tuning and interpretability, through derivation exercises and a scored quiz.
In simple words
Work through the derivations on paper first, then take the quiz. Each answer is explained. Anything you miss points to the topic to revisit before moving on to unsupervised learning.
Why it matters
These ideas recur in every advanced module: gradient descent and log loss train neural networks, regularisation becomes weight decay, and evaluation and leakage apply to every LLM system you will build.
How it works
Derivation exercises
- Derive the normal equation from J(w) = ‖y − Xw‖².
- Show that the gradient of log loss for logistic regression is Xᵀ(p − y)/n.
- Derive the ridge solution (XᵀX + λI)⁻¹Xᵀy and explain why it is always invertible for λ > 0.
- Compute the Gini decrease of a split from 8/2 into 6/0 and 2/2.
- Show that the cost-optimal threshold is c_FP / (c_FP + c_FN) for calibrated probabilities.
Common mistakes
- Tuning on the test set, or preprocessing before the split.
- Reporting accuracy on imbalanced data.
- Reading feature importance as causation.
Check yourself
Summary
- You can derive, implement, evaluate and tune the main supervised algorithms.
- You can design honest validation and choose metrics and thresholds from costs.
- Next: unsupervised learning and applied machine learning.
Finished reading?