Struct Kernel 

pub struct Kernel<'a> { /* private fields */ }

Kernel evaluator for training. Holds references to the dataset and precomputes x_square[i] = dot(x[i], x[i]) for RBF kernels.

The kernel_function method pointer pattern from C++ is replaced by a match on kernel_type — the branch predictor handles this efficiently since the type doesn’t change during training.

Implementations

impl<'a> Kernel<'a>

pub fn new(x: &'a [Vec<SvmNode>], param: &SvmParameter) -> Self

Create a new kernel evaluator for the given dataset and parameters.

pub fn evaluate(&self, i: usize, j: usize) -> f64

Evaluate K(x[i], x[j]) using precomputed data where possible.

pub fn swap_x_square(&mut self, i: usize, j: usize)

Swap indices i and j (used by the solver during working-set selection).

Note: this requires mutable access to the underlying data. Since the C++ code uses const_cast-style tricks, we handle this by requiring the caller to manage a mutable index mapping. The Kernel struct itself doesn’t own the data.