Skip to content

Index

This namespace contains the implementations of various restart mechanisms

IPOP

Bases: ModifyingRestart

Source code in evotorch/algorithms/restarter/modify_restart.py 
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
class IPOP(ModifyingRestart):
    def __init__(
        self,
        problem: Problem,
        algorithm_class: Type[SearchAlgorithm],
        algorithm_args: dict = {},
        min_fitness_stdev: float = 1e-9,
        popsize_multiplier: float = 2,
    ):
        """IPOP restart, terminates algorithm when minimum standard deviation in fitness values is hit, multiplies the population size in that case
        References:
            Glasmachers, Tobias, and Oswin Krause.
            "The hessian estimation evolution strategy."
            PPSN 2020
        Args:
            problem (Problem): A Problem to solve
            algorithm_class (Type[SearchAlgorithm]): The class of the search algorithm to restart
            algorithm_args (dict): Arguments to pass to the search algorithm on restart
            min_fitness_stdev (float): The minimum standard deviation in fitnesses; going below this will trigger a restart
            popsize_multiplier (float): A multiplier on the population size within algorithm_args
        """
        super().__init__(problem, algorithm_class, algorithm_args)

        self.min_fitness_stdev = min_fitness_stdev
        self.popsize_multiplier = popsize_multiplier

    def _search_algorithm_terminated(self) -> bool:
        # Additional check on standard deviation of fitnesses of population
        if self.search_algorithm.population.evals.std() < self.min_fitness_stdev:
            return True
        return super()._search_algorithm_terminated()

    def _modify_algorithm_args(self) -> None:
        # Only modify arguments if this isn't the first restart
        if self.num_restarts >= 1:
            new_args = deepcopy(self._algorithm_args)
            # Multiply population size
            new_args["popsize"] = int(self.popsize_multiplier * len(self.search_algorithm.population))
            self._algorithm_args = new_args

__init__(problem, algorithm_class, algorithm_args={}, min_fitness_stdev=1e-09, popsize_multiplier=2)

IPOP restart, terminates algorithm when minimum standard deviation in fitness values is hit, multiplies the population size in that case References: Glasmachers, Tobias, and Oswin Krause. "The hessian estimation evolution strategy." PPSN 2020 Args: problem (Problem): A Problem to solve algorithm_class (Type[SearchAlgorithm]): The class of the search algorithm to restart algorithm_args (dict): Arguments to pass to the search algorithm on restart min_fitness_stdev (float): The minimum standard deviation in fitnesses; going below this will trigger a restart popsize_multiplier (float): A multiplier on the population size within algorithm_args

Source code in evotorch/algorithms/restarter/modify_restart.py 
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
def __init__(
    self,
    problem: Problem,
    algorithm_class: Type[SearchAlgorithm],
    algorithm_args: dict = {},
    min_fitness_stdev: float = 1e-9,
    popsize_multiplier: float = 2,
):
    """IPOP restart, terminates algorithm when minimum standard deviation in fitness values is hit, multiplies the population size in that case
    References:
        Glasmachers, Tobias, and Oswin Krause.
        "The hessian estimation evolution strategy."
        PPSN 2020
    Args:
        problem (Problem): A Problem to solve
        algorithm_class (Type[SearchAlgorithm]): The class of the search algorithm to restart
        algorithm_args (dict): Arguments to pass to the search algorithm on restart
        min_fitness_stdev (float): The minimum standard deviation in fitnesses; going below this will trigger a restart
        popsize_multiplier (float): A multiplier on the population size within algorithm_args
    """
    super().__init__(problem, algorithm_class, algorithm_args)

    self.min_fitness_stdev = min_fitness_stdev
    self.popsize_multiplier = popsize_multiplier

ModifyingRestart

Bases: Restart

Source code in evotorch/algorithms/restarter/modify_restart.py 
23
24
25
26
27
28
29
30
31
class ModifyingRestart(Restart):
    def _modify_algorithm_args(self) -> None:
        """Modify the algorithm arguments on restart"""
        pass

    def _restart(self) -> None:
        """Restart the search algorithm"""
        self._modify_algorithm_args()
        return super()._restart()

Restart

Bases: SearchAlgorithm

Source code in evotorch/algorithms/restarter/restart.py 
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
class Restart(SearchAlgorithm):
    def __init__(
        self,
        problem: Problem,
        algorithm_class: Type[SearchAlgorithm],
        algorithm_args: dict = {},
        **kwargs: Any,
    ):
        """Base class for independent restarts methods
        Args:
            problem (Problem): A Problem to solve
            algorithm_class (Type[SearchAlgorithm]): The class of the search algorithm to restart
            algorithm_args (dict): Arguments to pass to the search algorithm on restart
        """

        SearchAlgorithm.__init__(
            self,
            problem,
            search_algorithm=self._get_sa_status,
            num_restarts=self._get_num_restarts,
            algorithm_terminated=self._search_algorithm_terminated,
            **kwargs,
        )

        self._algorithm_class = algorithm_class
        self._algorithm_args = algorithm_args

        self.num_restarts = 0
        self._restart()

    def _get_sa_status(self) -> dict:
        """Status dictionary of search algorithm"""
        return self.search_algorithm.status

    def _get_num_restarts(self) -> int:
        """Number of restarts (including the first start) so far"""
        return self.num_restarts

    def _restart(self) -> None:
        """Restart the search algorithm"""
        self.search_algorithm = self._algorithm_class(self._problem, **self._algorithm_args)
        self.num_restarts += 1

    def _search_algorithm_terminated(self) -> bool:
        """Boolean flag for search algorithm terminated"""
        return self.search_algorithm.is_terminated

    def _step(self):
        # Step the search algorithm
        self.search_algorithm.step()

        # If stepping the search algorithm has reached a terminal state, restart the search algorithm
        if self._search_algorithm_terminated():
            self._restart()

__init__(problem, algorithm_class, algorithm_args={}, **kwargs)

Base class for independent restarts methods Args: problem (Problem): A Problem to solve algorithm_class (Type[SearchAlgorithm]): The class of the search algorithm to restart algorithm_args (dict): Arguments to pass to the search algorithm on restart

Source code in evotorch/algorithms/restarter/restart.py 
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
def __init__(
    self,
    problem: Problem,
    algorithm_class: Type[SearchAlgorithm],
    algorithm_args: dict = {},
    **kwargs: Any,
):
    """Base class for independent restarts methods
    Args:
        problem (Problem): A Problem to solve
        algorithm_class (Type[SearchAlgorithm]): The class of the search algorithm to restart
        algorithm_args (dict): Arguments to pass to the search algorithm on restart
    """

    SearchAlgorithm.__init__(
        self,
        problem,
        search_algorithm=self._get_sa_status,
        num_restarts=self._get_num_restarts,
        algorithm_terminated=self._search_algorithm_terminated,
        **kwargs,
    )

    self._algorithm_class = algorithm_class
    self._algorithm_args = algorithm_args

    self.num_restarts = 0
    self._restart()