-
Notifications
You must be signed in to change notification settings - Fork 24
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
* local optimizer. base version * Merged with current main + added some comments * fixed names * fixed names in solver.py
- Loading branch information
1 parent
2639bb0
commit e0286a1
Showing
4 changed files
with
194 additions
and
67 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,150 @@ | ||
| import sys | ||
| from typing import List | ||
| from collections.abc import Callable | ||
|
|
||
| import scipy | ||
|
|
||
| from iOpt.method.optim_task import OptimizationTask | ||
| from iOpt.trial import Point, FunctionValue, FunctionType | ||
|
|
||
|
|
||
| class LocalTaskWrapper: | ||
| """ | ||
| Класс LocalTaskWrapper (название временное) оборачивает вычисление функции для дальнейшего применения локальных методов. | ||
| """ | ||
|
|
||
| def __init__(self, task: OptimizationTask, discrete_variables=None, max_calcs=-1): | ||
| self.discrete_variables = discrete_variables | ||
| self.task = task | ||
| self.calcs_count = 0 | ||
| self.max_calcs = max_calcs # В globalizer используется именно ограничение по количеству вычислений функции | ||
|
|
||
| def evaluate_function(self, y: List[float]) -> float: | ||
| """ | ||
| Метод вычисляет значение целевой функции | ||
| :param y: Точка в которой нужно вычислить значение функции | ||
| :return: возвращает значение целевой функции или | ||
| sys.float_info.max, если: | ||
| точка лежит за областью поиска | ||
| ИЛИ не были выполнены ограничения | ||
| ИЛИ было выкинуто исключение (функция не может быть посчитана в этой точке) | ||
| ИЛИ число вычислений превысило лимит (если он задан) | ||
| """ | ||
| point = Point(y, self.discrete_variables) | ||
| function_value = FunctionValue(FunctionType.OBJECTIV) | ||
| if self.max_calcs != -1 and self.calcs_count >= self.max_calcs: | ||
| function_value.value = sys.float_info.max | ||
| return function_value.value | ||
| for i in range(self.task.problem.dimension): | ||
| if (y[i] < self.task.problem.lower_bound_of_float_variables[i]) \ | ||
| or (y[i] > self.task.problem.upper_bound_of_float_variables[i]): | ||
| function_value.value = sys.float_info.max | ||
| return function_value.value | ||
|
|
||
| self.calcs_count += 1 | ||
| try: | ||
| for i in range(self.task.problem.number_of_constraints): | ||
| function_constraint_value = FunctionValue(FunctionType.CONSTRAINT, i) | ||
| function_constraint_value = self.task.problem.calculate(point, function_constraint_value) | ||
| if function_constraint_value.value > 0: | ||
| function_value.value = sys.float_info.max | ||
| return function_value.value | ||
|
|
||
| function_value = self.task.problem.calculate(point, function_value) | ||
| except Exception: | ||
| function_value.value = sys.float_info.max | ||
|
|
||
| return function_value.value | ||
|
|
||
|
|
||
| class HookeJeevesOptimizer: | ||
| """ | ||
| Класс HookeJeevesOptimizer реализует метод Хука-Дживса. | ||
| """ | ||
| def __init__(self, func: Callable[[List[float]], float], start_point: List[float], | ||
| step_mult: float, eps: float, max_iter: float): | ||
| self.nfev = 0 | ||
| self.cur_point = None | ||
| self.minf = None | ||
| self.best_point = None | ||
| self.pr_resdir = None | ||
| self.cur_resdir = None | ||
| self.dim = len(start_point) | ||
| self.f = func | ||
| self.start_point = start_point | ||
| self.max_iter = max_iter | ||
| self.eps = min(eps, 0.0001) | ||
| self.step = self.eps * 2 | ||
| self.step_mult = step_mult | ||
|
|
||
| def minimize(self) -> List[float]: | ||
| need_restart: bool = True | ||
| # self.best_point = self.start_point | ||
| # self.minf = self.f(self.start_point) | ||
| k, i, curr_f = 0, 0, 0.0 | ||
| while i < self.max_iter: | ||
| i += 1 | ||
| if need_restart: | ||
| k = 0 | ||
| self.cur_point = self.start_point.copy() | ||
| self.cur_resdir = self.start_point.copy() | ||
| curr_f = self.f(self.cur_point) | ||
| need_restart = False | ||
|
|
||
| self.pr_resdir = [el for el in self.cur_resdir] | ||
| self.cur_resdir = [el for el in self.cur_point] | ||
| next_f_value = self._make_research(self.cur_resdir) | ||
|
|
||
| if curr_f > next_f_value: | ||
| self._do_step() | ||
| k += 1 | ||
| curr_f = next_f_value | ||
| elif self.step > self.eps: | ||
| if k != 0: | ||
| self.start_point = self.pr_resdir.copy() | ||
| else: | ||
| self.step /= self.step_mult | ||
| need_restart = True | ||
| else: | ||
| break | ||
|
|
||
| return self.pr_resdir | ||
|
|
||
| def _make_research(self, point) -> float: | ||
| best_value = self.f(point) # в globalizer сделано так, хотя это значение уже было вычислено... | ||
|
|
||
| for i in range(self.dim): | ||
| point[i] += self.step | ||
| right_f_val = self.f(point) | ||
| if right_f_val > best_value: | ||
| point[i] -= 2 * self.step | ||
| left_f_val = self.f(point) | ||
|
|
||
| if left_f_val > best_value: | ||
| point[i] += self.step | ||
| else: | ||
| best_value = left_f_val | ||
|
|
||
| else: | ||
| best_value = right_f_val | ||
|
|
||
| return best_value | ||
|
|
||
| def _do_step(self) -> None: | ||
| for i in range(self.dim): | ||
| self.cur_point[i] = (1 + self.step_mult) * self.cur_resdir[i] - self.step_mult * self.pr_resdir[i] | ||
|
|
||
|
|
||
| def local_optimize(task: OptimizationTask, method, start_point: Point, args: dict, max_calcs: int = -1) -> dict: | ||
|
|
||
| local_task = LocalTaskWrapper(task=task, discrete_variables=start_point.discrete_variables, max_calcs=max_calcs) | ||
| if method == 'Hooke-Jeeves': | ||
| best_point = HookeJeevesOptimizer(local_task.evaluate_function, start_point.float_variables.copy(), | ||
| **args).minimize() | ||
| else: | ||
| best_point = scipy.optimize.minimize(local_task.evaluate_function, x0=start_point.float_variables.copy(), | ||
| method=method, **args).x | ||
|
|
||
| return {"x": best_point, "fev": local_task.calcs_count} |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.