Module classification 

Classification (ISL ch 4) — generative and instance-based classifiers.

• knn — k-nearest-neighbours (lazy, AllPairs).
• naive_bayes — Gaussian naive Bayes (generative, Reduction).

LDA / QDA (shared / per-class Gaussian discriminants) land here next, on the linear_algebra covariance solve (build order in stats_plan.md).

Re-exports

pub use discriminant::LdaModel;

pub use discriminant::QdaModel;

pub use knn::KnnClassifier;

pub use naive_bayes::GaussianNb;

pub use svm::Kernel;

pub use svm::Svm;

pub use svm_multiclass::MulticlassSvm;

Modules

discriminant

Discriminant analysis (ISL ch 4.4) — LDA (shared covariance ⇒ linear boundary) and QDA (per-class covariance ⇒ quadratic boundary). Both are Gaussian generative classifiers; the covariance inverse / log-determinant come from linear_algebra::cholesky (no new solver). Kernel-class DenseLinear.

knn

k-Nearest-Neighbours classifier (ISL ch 2/4) — a lazy learner: store the training set, classify a query by majority vote of its k nearest neighbours (squared Euclidean). Kernel-class AllPairs (query↔train distances).

naive_bayes

Gaussian Naive Bayes (ISL ch 4.4.4) — a generative classifier assuming the features are conditionally independent Gaussians given the class. Fit per-class priors and per-feature mean/variance (reusing statistics::descriptive); classify by the argmax log-posterior. Kernel-class Reduction.

svm

Support Vector Machine (ISL ch 9, PRML ch 7) — soft-margin binary classifier trained by simplified Sequential Minimal Optimization (SMO) on the dual, with a linear or RBF (Gaussian) kernel. Kernel SVM separates classes a linear boundary cannot. Labels are boolean (true = +1, false = −1). Kernel-class DenseLinear (the kernel matrix) + Divergent (the SMO working-set loop) → CPU here.

svm_multiclass

Multiclass SVM by one-vs-rest (ISL ch 9.4.2) — train one binary SVM per class (that class vs. all others) and predict the class whose decision value is largest. Reuses the binary super::svm (no duplicated SMO).