reconstruct_hessian_central_diff

scirs2_optimize::sparse_numdiff::compression

Function reconstruct_hessian_central_diff 

Source 
pub fn reconstruct_hessian_central_diff(
    gradients_forward: &ArrayView2<'_, f64>,
    gradients_backward: &ArrayView2<'_, f64>,
    p: &ArrayView2<'_, f64>,
    sparsity: &CsrArray<f64>,
    h: f64,
) -> Result<CsrArray<f64>, OptimizeError>
Expand description

Reconstructs a sparse Hessian from compressed gradient evaluations using central differences

Takes compressed gradient evaluations and reconstructs the symmetric sparse Hessian. The reconstruction assumes that gradients were computed using central differences with the compressed perturbation vectors.

§Arguments

  • gradients_forward - Forward difference gradients (n x num_colors)
  • gradients_backward - Backward difference gradients (n x num_colors)
  • p - Compression matrix (n x num_colors)
  • sparsity - Original Hessian sparsity pattern
  • h - Step size used in finite differences

§Returns

  • Reconstructed sparse Hessian