"""
Ready-to-run Evolution Strategy wrappers.
"""
from __future__ import annotations
from typing import Optional
import numpy as np
from ..encoding import Encoding
from ..objective_function import ObjectiveFunc
from ..initializers import UniformInitializer, PermInitializer
from ..encodings import TypeCastEncoding
from ..strategies import ES
from ..algorithms import Algorithm
from ..operators import create_operator
from ..survivor_selection import create_survivor_selection
from ..utils import RNGLike, check_random_state
[docs]
def evolution_strategy_binary(
objfunc: ObjectiveFunc,
mutated_bits: int = 1,
population_size: int = 100,
offspring_size: int = 500,
elitist: bool = False,
encoding: Optional[Encoding] = None,
random_state: Optional[RNGLike] = None,
**kwargs,
) -> Algorithm:
"""Evolution Strategy for binary-coded vectors.
Parameters
----------
objfunc : ObjectiveFunc
The objective function to optimise.
mutated_bits : int, optional
Number of bits flipped per mutation (default 1).
population_size : int, optional
Population size (default 100).
offspring_size : int, optional
Number of offspring per generation (default 500).
elitist : bool, optional
If ``True``, use (μ+λ) selection; otherwise (μ,λ).
encoding : Encoding, optional
Encoding; defaults to :class:`TypeCastEncoding` (int → bool).
random_state : RNGLike, optional
Random seed or generator.
**kwargs
Forwarded to :class:`Algorithm`.
"""
random_state = check_random_state(random_state)
encoding = TypeCastEncoding(int, bool) if encoding is None else encoding
pop_initializer = UniformInitializer(
objfunc.dimension, 0, 1, population_size=population_size, dtype=np.uint8, encoding=encoding, random_state=random_state
)
mutation_op = create_operator("mutation.bitflip", N=mutated_bits, random_state=random_state)
method = "keep_best" if elitist else "keep_offspring"
survivor_sel = create_survivor_selection(method, random_state=random_state)
search_strat = ES(
initializer=pop_initializer, mutation_op=mutation_op, survivor_sel=survivor_sel, offspring_size=offspring_size, random_state=random_state
)
return Algorithm(objfunc, search_strat, **kwargs)
[docs]
def evolution_strategy_permutation(
objfunc: ObjectiveFunc,
swapped_positions: int = 2,
population_size: int = 100,
offspring_size: int = 500,
elitist: bool = False,
encoding: Optional[Encoding] = None,
random_state: Optional[RNGLike] = None,
**kwargs,
) -> Algorithm:
"""Evolution Strategy for permutation-coded vectors.
Parameters
----------
objfunc : ObjectiveFunc
The objective function to optimise.
swapped_positions : int, optional
Number of positions swapped per mutation (default 2).
population_size : int, optional
Population size (default 100).
offspring_size : int, optional
Number of offspring per generation (default 500).
elitist : bool, optional
If ``True``, use (μ+λ) selection; otherwise (μ,λ).
encoding : Encoding, optional
Encoding applied to the genotype.
random_state : RNGLike, optional
Random seed or generator.
**kwargs
Forwarded to :class:`Algorithm`.
"""
random_state = check_random_state(random_state)
pop_initializer = PermInitializer(objfunc.dimension, population_size=population_size, encoding=encoding, random_state=random_state)
mutation_op = create_operator("permutation.swap", N=swapped_positions, random_state=random_state)
method = "keep_best" if elitist else "keep_offspring"
survivor_sel = create_survivor_selection(method, random_state=random_state)
search_strat = ES(
initializer=pop_initializer, mutation_op=mutation_op, survivor_sel=survivor_sel, offspring_size=offspring_size, random_state=random_state
)
return Algorithm(objfunc, search_strat, **kwargs)
[docs]
def evolution_strategy_discrete(
objfunc: ObjectiveFunc,
resampled_components: int = 1,
population_size: int = 100,
offspring_size: int = 500,
elitist: bool = False,
encoding: Optional[Encoding] = None,
random_state: Optional[RNGLike] = None,
**kwargs,
) -> Algorithm:
"""Evolution Strategy for integer-coded vectors.
Parameters
----------
objfunc : ObjectiveFunc
The objective function to optimise.
resampled_components : int, optional
Number of components resampled per mutation (default 1).
population_size : int, optional
Population size (default 100).
offspring_size : int, optional
Number of offspring per generation (default 500).
elitist : bool, optional
If ``True``, use (μ+λ) selection; otherwise (μ,λ).
encoding : Encoding, optional
Encoding applied to the genotype.
random_state : RNGLike, optional
Random seed or generator.
**kwargs
Forwarded to :class:`Algorithm`.
"""
random_state = check_random_state(random_state)
pop_initializer = UniformInitializer(
objfunc.dimension,
objfunc.lower_bound,
objfunc.upper_bound,
population_size=population_size,
dtype=int,
encoding=encoding,
random_state=random_state,
)
mutation_op = create_operator("random.reset", n=resampled_components, random_state=random_state)
method = "keep_best" if elitist else "keep_offspring"
survivor_sel = create_survivor_selection(method, random_state=random_state)
search_strat = ES(
initializer=pop_initializer, mutation_op=mutation_op, survivor_sel=survivor_sel, offspring_size=offspring_size, random_state=random_state
)
return Algorithm(objfunc, search_strat, **kwargs)
[docs]
def evolution_strategy_real(
objfunc: ObjectiveFunc,
mutation_strength: float = 1e-2,
mutated_components: int = 1,
population_size: int = 100,
offspring_size: int = 500,
elitist: bool = False,
encoding: Optional[Encoding] = None,
random_state: Optional[RNGLike] = None,
**kwargs,
) -> Algorithm:
"""Evolution Strategy for integer-coded vectors.
Parameters
----------
objfunc : ObjectiveFunc
The objective function to optimise.
resampled_components : int, optional
Number of components resampled per mutation (default 1).
population_size : int, optional
Population size (default 100).
offspring_size : int, optional
Number of offspring per generation (default 500).
elitist : bool, optional
If ``True``, use (μ+λ) selection; otherwise (μ,λ).
encoding : Encoding, optional
Encoding applied to the genotype.
random_state : RNGLike, optional
Random seed or generator.
**kwargs
Forwarded to :class:`Algorithm`.
"""
random_state = check_random_state(random_state)
pop_initializer = UniformInitializer(
objfunc.dimension,
objfunc.lower_bound,
objfunc.upper_bound,
population_size=population_size,
dtype=float,
encoding=encoding,
random_state=random_state,
)
mutation_op = create_operator("mutation.gaussian_mutation", F=mutation_strength, N=mutated_components, random_state=random_state)
method = "keep_best" if elitist else "keep_offspring"
survivor_sel = create_survivor_selection(method, random_state=random_state)
search_strat = ES(
initializer=pop_initializer, mutation_op=mutation_op, survivor_sel=survivor_sel, offspring_size=offspring_size, random_state=random_state
)
return Algorithm(objfunc, search_strat, **kwargs)