classical-mlsupervisednon-parametric
k-Nearest neighbors
Instance-based learning — no training phase, just a distance metric and a vote.
Уровни глубины
L0Intro~1ч
Classifies a point by majority of its k nearest labelled points.
L1Basics~6ч
Implements brute-force kNN; analyses curse of dimensionality; chooses metric.
L2Working~12ч
Uses KD-tree, Ball-tree, HNSW/Faiss for approximate NN at scale.
L3Advanced~20ч
Distance metric learning; local methods for regression (kernel smoothers).
L4Research~40ч
Theoretical guarantees of kNN in high dimension; connections to contrastive learning.
Ресурсы
L0 — Intro
L1 — Basics
L2 — Working