example.example_experiment_non_anytime

module documentation

An example experiment for benchmarking non-anytime optimization algorithms with restarts.

Non-anytime algorithms are those whose parameters depend not only on the dimensionality of the problem, but also on the budget of evaluations. Benchmarking such algorithms therefore entails running experiments on a test suite using various (increasing) budgets. These budgets can/should be provided by the user. If the user does not specify a budget list, a default budget list is used.

This code builds upon the example experiment for beginners and provides only simplistic progress information.

To apply the code to a different solver, add an interface to function fmin.

Function default_budget_list Produces a budget list with at most num different increasing budgets within [1, max_budget] that are equally spaced in the logarithmic space.
Function fmin Invokes solver on problem with budget and initial solution x0. Returns the final solution.
Function main Undocumented
def default_budget_list(max_budget=10, num=50):

Produces a budget list with at most num different increasing budgets within [1, max_budget] that are equally spaced in the logarithmic space.

def fmin(problem, x0, solver, budget):

Invokes solver on problem with budget and initial solution x0. Returns the final solution.

Add here the interface to your solver.

def main():

Undocumented