1.3. The Learning Paradigm

💡 First Principle: The presence or absence of labels in your training data determines which algorithms you can use. Labels present means supervised learning. No labels means unsupervised learning. This single distinction eliminates half the options on any ML-related exam question.

What breaks without understanding this: You'll see questions asking "Which technique groups customers by similar attributes without predefined categories?" Without the supervised/unsupervised framework, you might guess classification. But "without predefined categories" means no labels—that's clustering, an unsupervised technique. Miss this distinction and you miss the question.

Imagine a teacher grading exams. Supervised learning is like a teacher with an answer key—every question has a known correct answer to compare against. Unsupervised learning is like a teacher asked to group students by "similar learning styles" with no predefined categories—patterns must be discovered, not verified. For instance, consider a retail scenario: predicting whether a customer will buy (yes/no labels) is supervised classification, while grouping customers by purchasing behavior (no labels) is unsupervised clustering. The exam tests whether you can identify which situation applies.