Skip to content
learn
  1. Learn
  2. Machine Learning
  3. Regression

ML-M03

Regression

Predicting numbers: linear regression, how it is solved, gradient descent, curves, regularization and the metrics that judge it.

6 lessonsabout 80 min

Regression is the most direct form of machine learning: given some features, predict a number, such as a price, a delivery time or tomorrow's demand. This module starts with the straight line and builds out every idea that the rest of the track reuses: a loss that measures mistakes, two ways to minimise it (exact algebra and gradient descent), how adding curved features leads to overfitting, how regularization keeps a model in check, and how to choose a metric that matches what an error really costs.

You will drag lines and descend loss surfaces by hand before seeing the formulas, and every result is checked with a few lines of NumPy or scikit-learn.

Lessons

  1. ML-03.01 Linear regression intuition Linear regression fits a straight line, or a flat plane in many dimensions, through data: prediction equals weights times features plus a bias. Beginner 14 min
  2. ML-03.02 Solving linear regression: normal equation and gradient descent Linear regression can be solved exactly with the normal equation, which computes the best weights with matrix algebra in one step. Intermediate 15 min
  3. ML-03.03 Gradient descent, the engine of ML Gradient descent minimizes a loss by repeatedly stepping in the direction opposite to the gradient: new weight equals old weight minus learning rate times gradient. Beginner 18 min
  4. ML-03.04 Polynomial regression and basis functions A linear model can fit curves if you give it non-linear features, such as x squared and x cubed. Intermediate 10 min
  5. ML-03.05 Regularization: ridge, lasso, elastic net Regularization adds a penalty on weight size to the loss so the model prefers simpler solutions. Intermediate 14 min
  6. ML-03.06 Regression metrics MAE averages absolute errors and is easy to explain; MSE and RMSE square errors, punishing large mistakes more; MAPE expresses errors as percentages but breaks near zero; R squared measures the share of variance explained. Beginner 10 min

Start with lesson 1