simopt.solvers.utils

Utility functions for optimization solvers.

Module Contents

simopt.solvers.utils.finite_diff(solver: simopt.base.Solver, new_solution: simopt.base.Solution, bounds_check: numpy.ndarray, problem: simopt.base.Problem, stepsize: float, r: int) → numpy.ndarray

Compute the finite difference approximation of the gradient for a solution.

Parameters:

• solver (Solver) – The solver instance used to create new solutions.

• new_solution (Solution) – The current solution to perturb.

• bounds_check (np.ndarray) – Array indicating which perturbation method to use per dimension.

• problem (Problem) – The problem instance providing bounds and function evaluations.

• stepsize (float) – The step size used for finite difference calculations.

• r (int) – The number of replications used for each function evaluation.

Returns:

The approximated gradient of the function at the given solution.

Return type:

np.ndarray

simopt.solvers.utils.bfgs_hessian_approx(solver: simopt.base.Solver, new_solution: simopt.base.Solution, bounds_check: numpy.ndarray, problem: simopt.base.Problem, r: int) → numpy.ndarray

Approximate Hessian using BFGS.

This method applies BFGS updates to build or refine a Hessian approximation.

Parameters:

• solver (Solver) – The solver instance used to create new solutions.

• new_solution (Solution) – The solution at which derivatives are computed.

• bounds_check (np.ndarray) – Boolean mask indicating which variables are within bounds and eligible for perturbation.

• problem (Problem) – The simulation-optimization problem being solved.

• r (int) – Number of replications used when estimating gradients.

Returns:

Hessian approximation (updated via BFGS).

Return type:

np.ndarray