Primers • K-Nearest Neighbors (kNN)

first video in a while, suffering from low blood platelets currently, apologies beforehand for any inconvenience k-Nearest Neighbours. This is Episode 01 of the basics-to-frontier series on this channel. · Why kNN is supervised, non-parametric, lazy and instance-based · The two phases: an empty training phase, and a four-step prediction phase · Euclidean, Manhattan, Minkowski and Hamming distance , and the geometry of each · How the choice of metric alone can change which point is "nearest" · Choosing k: overfitting, underfitting, and cross-validation with real error curves · Distance weighting, and when it actually changes the answer · A complete worked example: four points, one query, k = 1, 3 and 4 · Seven ways to avoid committing to a fixed k, and what each one costs · kNN vs k-means: same letter, completely different job Everything numerical on screen is computed, not typed by hand — the decision boundaries, the roughness percentages, the error-vs-k curves and the LOF scores are all produced by the same code that renders the animation. Animations built with Manim Community Edition. Corrections and questions go in the comments; I read them. knowledge should be open source.