Skip to main content
Code ScholarsCodeScholars
← AI & Machine Learning Study Guide

Unit 3 · Heavy

Supervised Learning

The algorithms themselves, treated as tools with known behavior rather than as the point of the course. What matters is knowing what each assumes and how each fails.

What a strong answer looks like

A strong Unit 3 answer names what the model assumes about the data and how you would detect that the assumption is wrong.

Practice this unitCourse overview

Topics in this unit

1

Linear Models

Know

A coefficient is the expected change in output per unit change in that feature, holding others fixed. Regularisation penalises large coefficients to discourage fitting noise.

Apply

Interpret coefficients in the units of the problem, and use regularisation when features are many or correlated.

Watch out

Interpreting a coefficient causally. It describes an association under the model, not a cause.

Study move

Interpret two coefficients from a fitted model in plain language with units.

2

Logistic Regression and Thresholds

Know

Logistic regression outputs a probability that is then thresholded into a class. Moving the threshold trades recall against precision.

Apply

Choose the threshold from the relative cost of the two error types, not by default.

Watch out

Treating 0.5 as fixed. In screening or fraud the right threshold is rarely a half.

Study move

Say what happens to precision and recall as the threshold falls.

3

Decision Trees

Know

A tree splits on features to reach a leaf, which makes it interpretable. Unconstrained depth lets it isolate every training point and memorise noise.

Apply

Constrain depth or minimum leaf size, and use the tree structure to explain a decision.

Watch out

An unpruned tree scoring perfectly on training data and poorly on anything else.

Study move

Describe what a very deep tree has learned that a shallow one has not.

4

Nearest Neighbours

Know

Classification by proximity requires scaled features, and the neighbour count controls the smoothness of the boundary.

Apply

Scale first, then tune the neighbour count, trading variance against bias.

Watch out

Running the method on unscaled features, where the largest-range feature dominates the distance.

Study move

State what happens to the boundary as the neighbour count grows.

Emphasized in this unit

Connections and techniques that receive extra attention in this unit.

  • • Interpreting coefficients with units and without causal claims
  • • Choosing a threshold from error costs
  • • Constraining tree depth deliberately

Varies by course

Related topics some schools attach to this unit and others leave out. Covered on request rather than assumed.

  • Ensembles. Random forests and boosting appear in some sections.
  • Support vector machines. Covered in some tracks only.

Mastery checklist

  • • Interpret a coefficient in problem units.
  • • Say how threshold changes affect precision and recall.
  • • Explain why an unpruned tree overfits.
  • • Explain why nearest neighbours needs scaling.

Check yourself

  • • Why is a coefficient not a causal claim?
  • • When is a 0.5 threshold the wrong choice?
  • • What does raising the neighbour count do to the boundary?

Modeling drill

For a fraud detector where a missed fraud costs far more than a false alarm, choose a model and a threshold, and justify both.

CoefficientRegularisationThresholdDecision treePruningk-NNLoss function

Previous

Unit 2: Data Preparation and Features

Next

Unit 4: Model Evaluation