6 lessons
Linear regression, from a straight line to real math
Building the simplest predictive model there is, and deriving exactly how it learns using the mean squared error and gradient descent.
Logistic regression, predicting a category instead of a number
Reusing the linear regression machinery, adding the sigmoid function, and deriving the log loss that trains a classifier.
Decision trees, learning by asking questions
A completely different kind of model with no weights or gradients at all — just a sequence of yes/no questions, chosen using entropy and information gain.
K-nearest neighbors, predicting by asking your neighbors
A model with no training step at all, no weights, and no gradients — just the distance formula and a vote, and why that makes it both simple and slow.
Building a real Naive Bayes spam classifier
Turning the Bayes' theorem math from the foundations module into a complete, working text classifier, end to end.
Support vector machines, the widest possible margin
A classifier built around a genuinely different goal - not just separating two classes, but separating them with as much breathing room as possible.