Model Building & Algorithm Learning

Regression Playground

Explore how regression models fit signal, respond to noise and outliers, and trade flexibility against regularisation.

7 · Regression 8 · Clustering 9 · Decision boundaries 10 · Neural networks 11 · Gradient descent 12 · Ensembles 13 · Step-by-step

Lab concept guide

What to observe while you experiment

Regression predicts a quantitative target. Model families differ in the functions they can represent and in how regularisation or neighbourhood assumptions trade bias against variance; residuals reveal structure that a headline score can hide.

MechanismFit one regressor, inspect the fitted function and residuals, then change noise/outliers/complexity while keeping the evaluation protocol fixed.

Failure modeJudging fit only from a smooth-looking curve or training R² without inspecting held-out residuals and extrapolation.

VerificationHand-check a few residuals and one metric component, then compare training and validation error as flexibility/regularisation changes.

Experiment deliberately

Use linear and nonlinear data with an outlier. Predict which model bends toward the outlier and how regularisation/flexibility should change residual patterns.

DatasetModelPolynomial degree2Noise7Regularisation λ0.08Seed

Interact with the observations. Add a custom point with x/y values, inject an outlier, remove your last custom point, or regenerate the synthetic sample.

x y

Ready.

Observed data and fitted function

Compare the model curve with the underlying observations.

Residual diagnostics

Residuals should not show strong systematic structure when the functional form is adequate.