Struct DifferentiableLP 

pub struct DifferentiableLP {
    pub c: Vec<f64>,
    pub a_eq: Vec<Vec<f64>>,
    pub b_eq: Vec<f64>,
    pub g: Vec<Vec<f64>>,
    pub h: Vec<f64>,
}

A differentiable LP layer.

Fields

c: Vec<f64>

Objective coefficient vector c (n).

a_eq: Vec<Vec<f64>>

Equality constraint matrix A (p×n).

b_eq: Vec<f64>

Equality constraint rhs b (p).

g: Vec<Vec<f64>>

Inequality constraint matrix G (m×n): Gx ≤ h.

h: Vec<f64>

Inequality constraint rhs h (m).

Implementations

impl DifferentiableLP

pub fn new( c: Vec<f64>, a_eq: Vec<Vec<f64>>, b_eq: Vec<f64>, g: Vec<Vec<f64>>, h: Vec<f64>, ) -> OptimizeResult<Self>

Create a new differentiable LP.

pub fn n(&self) -> usize

Number of primal variables.

pub fn forward(&self, config: &DiffLPConfig) -> OptimizeResult<DiffLPResult>

Solve the LP (forward pass).

Internally converts to a QP with Q = εI (small regularization) and solves via interior-point method.

pub fn backward( &self, result: &DiffLPResult, dl_dx: &[f64], config: &DiffLPConfig, ) -> OptimizeResult<ImplicitGradient>

Backward pass: compute gradients of loss w.r.t. LP parameters.

Uses active-set implicit differentiation: at the LP optimum, only active inequality constraints matter, and we differentiate through the resulting equality system.

Trait Implementations

impl Clone for DifferentiableLP

fn clone(&self) -> DifferentiableLP

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for DifferentiableLP

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.