k-nearest neighbours: ask the closest points
Two labeled classes of points sit on a plane alongside one new, unlabeled point. For k = 1, 3, and 7, the animation highlights the k closest neighbors, tallies their class labels, and shows the majority vote assigning the new point's class. A closing note contrasts small k (sensitive, noisy) versus large k (smoother, more biased) choices. Useful for students and teachers introducing instance-based learning, where prediction relies directly on stored data rather than a trained parametric model.
16:9 · every frame verified for overlaps, spacing and edges before rendering
The prompt that made it
Two classes of points and a new point; for k = 1, 3 and 7 the k nearest are highlighted, counted and the majority decides, with a note on small versus large k. No training step: the data is the model.
Make the next one in this series
- Show how changing the distance metric (Euclidean vs Manhattan) changes which neighbors are chosen.
- Animate decision boundaries forming as k increases across the whole plane, not just one point.
- Compare KNN classification to KNN regression on a simple 1D dataset.
