Stop 05 · Machine Learning FundamentalsTopics · 1/12

Supervised vs Unsupervised Learning

In short

Before you pick any algorithm, ask one question: do I have the answers for my training data? That single question splits most of machine learning into two camps.

Supervised learning trains on labeled data to predict outcomes; unsupervised learning finds structure in unlabeled data — the first fork in the road for choosing an ML approach.

The blob studies a solved answer sheet on one side and sorts unlabelled spice jars into groups on the other.

Think of it like this

Solved papers vs a mixed-up spice box

Supervised learning is practising with last year's solved papers: every question comes with the right answer. Unsupervised learning is your mom's unlabelled spice box. Nobody tells you the names, but you can still group them by colour and smell.

Watch it happen

Do you have labels?

Supervised

Each row has an answer (label)
Predicts price, spam, churn
Scored against true answers

Unsupervised

Only features, no answers
Finds groups and patterns
Judged by how useful groups are

Have labels and a clear target? Go supervised. Only raw data and a question like 'what kinds of customers do we have?' Go unsupervised.

The big ideas

What you need to know

Labels are the answers

A label is the value you want to predict, like 'spam' or a house price. Supervised models learn the link from features to labels.

Two supervised flavours

Regression predicts a number (rent, temperature). Classification predicts a category (pass or fail, fraud or genuine).

Unsupervised finds structure

Clustering groups similar rows. Dimensionality reduction squeezes many columns into a few that keep most of the information.

Labels cost money

Someone has to tag every example by hand. That's why teams often cluster first, then label a smaller, smarter sample.

See it in code

Walk through the code

python

1from sklearn.datasets import load_iris2from sklearn.linear_model import LogisticRegression3from sklearn.cluster import KMeans4 5X, y = load_iris(return_X_y=True)6 7# Supervised: learn from features AND labels8clf = LogisticRegression(max_iter=200).fit(X, y)9print(clf.predict(X[:3]))   # [0 0 0]10 11# Unsupervised: only features, find groups12km = KMeans(n_clusters=3, n_init=10, random_state=0).fit(X)13print(km.labels_[:3])       # cluster ids, numbering is arbitrary

1/3X holds the measurements of each flower. y holds the species, which is the label.

Same flower data, two jobs. The classifier sees y and predicts species. KMeans never sees y and invents its own group numbers.

Save yourself the pain

Common mistakes

Oops

Thinking clustering will 'predict' the right category names for you.

Do this instead

Clusters are just groups with numbers. You still have to look inside each one and decide what it means.

Oops

Using a supervised model when you have only a handful of labelled rows.

Do this instead

Get more labels first, or start with clustering and labelling a small sample from each cluster.

Your mission

One dataset, two questions

~20 min

Quick check

A shop has a year of customer purchase data but no tags on the customers. They want to find shopper types. What fits best?

Go deeper

1List five real-world problems and classify each as supervised, unsupervised, or unclear.
2Learn the sub-categories: classification/regression under supervised, clustering/dimensionality reduction under unsupervised.
3Understand semi-supervised and self-supervised learning as the bridge to modern deep learning.