src.manfred.direct_search

Module Contents

Functions

do_manfred_direct_search(func, current_x, step_size, state, direction_window, bounds, mode, n_evaluations_per_x, batch_evaluator, batch_evaluator_options)	Search for better values along coordinates and 45 degree lines.
_determine_search_strategies(current_x, state, direction_window, mode)
_determine_strategies_from_residuals(current_x, state)
_determine_strategies_from_x_history(current_x, state, direction_window)
_get_direct_search_sample(current_x, step_size, search_strategies, bounds)
_bools_to_strategy(left, right)
_combine_strategies(resid, hist)

do_manfred_direct_search(func, current_x, step_size, state, direction_window, bounds, mode, n_evaluations_per_x, batch_evaluator, batch_evaluator_options)

Search for better values along coordinates and 45 degree lines.

For each parameter we try to find the most promising direction based on previous function evaluations and/or the sign of the residuals (if mode == “fast”) or search in both directions (if mode == “thorough”).

Moreover, we search over all combination of steps. This makes the step infeasible for high parameter dimensions but robust at a moderate cost for low parameter dimensions.

_determine_search_strategies(current_x, state, direction_window, mode)

_determine_strategies_from_residuals(current_x, state)

_determine_strategies_from_x_history(current_x, state, direction_window)

_get_direct_search_sample(current_x, step_size, search_strategies, bounds)

_bools_to_strategy(left, right)

_combine_strategies(resid, hist)