Detailed Description

Solver Policy 3: LU Decomposition.

This solver uses LU decomposition with partial pivoting to factorize the system matrix B = I + Sigma * W. This is the most numerically robust solver but also the slowest, as it does not exploit the positive definiteness of B that the Cholesky solvers assume.

This solver is used as a fallback when Solvers 1 and 2 fail due to numerical issues (e.g., near-singular Hessians).

Note: This solver corresponds to solver == 3 in the original code.

Definition at line 821 of file laplace_marginal_density_estimator.hpp.

#include <laplace_marginal_density_estimator.hpp>

Public Member Functions
template<typename NewtonStateT , typename LLFun , typename LLTupleArgs , typename CovarMat >
void	solve_step (NewtonStateT &state, const LLFun &ll_fun, const LLTupleArgs &ll_args, const CovarMat &covariance, int hessian_block_size, std::ostream *msgs)
	Perform one Newton step using LU decomposition solver.

double	compute_log_determinant () const
	Compute log determinant from LU factorization.

template<typename NewtonStateT >
auto	build_result (NewtonStateT &state, double log_det)
	Build the final result structure.

Public Attributes
Eigen::PartialPivLU< Eigen::MatrixXd >	lu
	LU factorization of B = I + Sigma * W.

Eigen::SparseMatrix< double >	W_full
	Full Hessian matrix from likelihood.

The documentation for this struct was generated from the following file:

stan/math/mix/functor/laplace_marginal_density_estimator.hpp

Detailed Description

Public Member Functions

Public Attributes