A4 · machine learning

Machine learning, you can poke

Each demo is a faithful, hands-on visualisation of one standard idea from the A4 syllabus. Every one runs the actual algorithm underneath, not a scripted animation.

• A4.3.2

  k-nearest neighbours

  Click points into the plane, drag k, and watch the decision boundary form from nothing but a vote of the nearest neighbours.

  Open→
• A4.3.1

  Linear regression and gradient descent

  Watch a line learn to fit a scatter by taking small steps downhill on its error, with the loss falling toward the exact best fit.

  Open→
• A4.3.4

  k-means clustering

  With no labels at all, watch k centres drift to the heart of each cluster as the algorithm assigns and recentres, step by step.

  Open→
• A4.3.8

  The perceptron

  Train one artificial neuron to learn a logic gate, watch its boundary line rotate into place, then meet the XOR limit that needs a hidden layer.

  Open→
• A4.3.7

  Genetic algorithm

  Evolve random text toward a target you type by selection, crossover and mutation, with the fittest climbing every generation.

  Open→
• A4.3.6

  Reinforcement learning

  An agent learns a path through a grid world by trial and reward; the values fill in, a policy emerges, and you set how much it explores.

  Open→
• A4.3.9

  Convolution and feature maps

  Watch a filter slide across an image to build a feature map, then pool it down: the two operations a convolutional layer is built from.

  Open→
• A4.3.3

  Overfitting and the metrics

  Slide a polynomial from too simple to too eager and watch train vs test error, then drive accuracy, precision, recall and F1 from one threshold.

  Open→
• A4.4.1

  Accuracy versus fairness

  Train a classifier on a biased loan dataset and watch a model stay highly accurate while quietly treating two groups very differently.

  Open→