Struct NlsSolution 

pub struct NlsSolution {
    pub theta: Vec<f64>,
    pub cost: f64,
    pub iters: usize,
    /* private fields */
}

The solved layer: the optimum, its cost, and the machinery for exact outer sensitivities.

Fields

theta: Vec<f64>

cost: f64

iters: usize

Implementations

impl NlsSolution

pub fn dtheta_dphi<R: Residual>( &self, res: &R, phi: &[f64], k: usize, ) -> Vec<f64>

Exact sensitivity dθ*/dφ_k by implicit differentiation: −(JᵀJ)⁻¹ Jᵀ ∂r/∂φ_k, with ∂r/∂φ_k from one dual pass seeded in the outer parameter.

pub fn backward<R: Residual>( &self, res: &R, phi: &[f64], dl_dtheta: &[f64], ) -> Vec<f64>

Chain an outer loss gradient through the layer: given ∂L/∂θ*, return ∂L/∂φ (one factored solve plus one dual pass per outer parameter).