Struct KernelSVMTrainer 

pub struct KernelSVMTrainer {
    pub n: usize,
    pub alphas: Vec<f64>,
    pub labels: Vec<f64>,
    pub bias: f64,
    pub c: f64,
}

Kernel SVM trainer using a simplified SMO algorithm.

Implements the Sequential Minimal Optimization (SMO) core update step.

Fields

n: usize

Number of training points.

alphas: Vec<f64>

Dual variables α_i ∈ [0, C].

labels: Vec<f64>

Labels y_i ∈ {-1, +1}.

bias: f64

Bias term b.

c: f64

Regularization parameter C (upper bound on α_i).

Implementations

impl KernelSVMTrainer

pub fn new(n: usize, labels: Vec<f64>, c: f64) -> Self

Create a new KernelSVMTrainer with zero alphas.

pub fn decision(&self, kernel_row: &[f64]) -> f64

Compute the SVM decision function f(x) = Σ α_i y_i k(x_i, x) + b.

pub fn smo_step(&mut self, i: usize, j: usize, k: &[Vec<f64>]) -> bool

One SMO update step on a pair (i, j) given kernel matrix K.

Returns true if a meaningful update was made.

pub fn generalization_bound(radius: f64, margin: f64, n: usize) -> f64

SVM generalization bound: ≤ R²/(γ² n) with margin γ.