for i, k in enumerate(params):
if i<self.n_cross or random.random() < self.cr:
# cfg = x1 + use_f*(x2 - x3)
cfg_params[k].set_linear(1.0, x1, use_f, x2, -use_f, x3)
pp.touch() # move to back of the line for next replacement
pp.candidate_replacement = self.driver.get_configuration(cfg)
self.limit = self.driver.objective.limit_from_config(pp.config)
return pp.candidate_replacement
def handle_requested_result(self, result):
'''called when new results are added'''
for p in self.population:
if p.candidate_replacement == result.configuration:
if self.objective.lt(p.candidate_replacement, p.config):
# candidate replacement was better, replace it!
p.config = p.candidate_replacement
log.info("better point")
p.candidate_replacement = None
class DifferentialEvolutionAlt(DifferentialEvolution):
def __init__(self, cr=0.2, **kwargs):
kwargs['cr'] = cr
super(DifferentialEvolutionAlt, self).__init__(**kwargs)
register(DifferentialEvolution(name='de'))
register(DifferentialEvolutionAlt())
for i, k in enumerate(params):
if i < self.n_cross or random.random() < self.cr:
# cfg = x1 + use_f*(x2 - x3)
cfg_params[k].set_linear(1.0, x1, use_f, x2, -use_f, x3)
return cfg
def handle_requested_result(self, result):
"""called when new results are added"""
for p in self.population:
if p.candidate_replacement == result.configuration:
if self.objective.lt(p.candidate_replacement, p.config):
# candidate replacement was better, replace it!
p.config = p.candidate_replacement
log.info('better point')
p.candidate_replacement = None
class DifferentialEvolutionAlt(DifferentialEvolution):
def __init__(self, cr=0.2, **kwargs):
kwargs['cr'] = cr
super(DifferentialEvolutionAlt, self).__init__(**kwargs)
register(DifferentialEvolution())
register(DifferentialEvolutionAlt())
register(
DifferentialEvolution(population_size=100,
cr=0.2,
name='DifferentialEvolution_20_100'))
return population
def select_parameters(self, params):
random.shuffle(params)
ret_list = params[:self.must_mutate_count]
for param in params[self.must_mutate_count:]:
if random.random() < self.mutation_rate:
ret_list.append(param)
return ret_list
def get_default_operator(self, param_type):
return {'op_name': 'op1_randomize', 'args': [], 'kwargs': {}}
register(
RandomThreeParentsComposableTechnique(name='ComposableDiffEvolution',
population_size=30))
register_generator(RandomThreeParentsComposableTechnique)
register_generator(GreedyComposableTechnique)
op_map = {}
ComposableEvolutionaryTechnique.add_to_map(op_map,
PermutationParameter,
"op3_cross",
xchoice='op3_cross_CX')
ComposableEvolutionaryTechnique.add_to_map(op_map,
"FloatArray",
"op3_cross",
strength=0.4)
register(
RandomThreeParentsComposableTechnique(name='ComposableDiffEvolutionCX',
params = self.manipulator.param_names(cfg, x1, x2, x3)
random.shuffle(params)
for i, k in enumerate(params):
if i < self.n_cross or random.random() < self.cr:
# cfg = x1 + use_f*(x2 - x3)
cfg_params[k].set_linear(1.0, x1, use_f, x2, -use_f, x3)
return cfg
def handle_requested_result(self, result):
"""called when new results are added"""
for p in self.population:
if p.candidate_replacement == result.configuration:
if self.objective.lt(p.candidate_replacement, p.config):
# candidate replacement was better, replace it!
p.config = p.candidate_replacement
log.info('better point')
p.candidate_replacement = None
class DifferentialEvolutionAlt(DifferentialEvolution):
def __init__(self, cr=0.2, **kwargs):
kwargs['cr'] = cr
super(DifferentialEvolutionAlt, self).__init__(**kwargs)
register(DifferentialEvolution())
register(DifferentialEvolutionAlt())
params = self.manipulator.param_names(cfg, x1, x2, x3)
random.shuffle(params)
for i, k in enumerate(params):
if i < self.n_cross or random.random() < self.cr:
# cfg = x1 + use_f*(x2 - x3)
cfg_params[k].set_linear(1.0, x1, use_f, x2, -use_f, x3)
return cfg
def handle_requested_result(self, result):
"""called when new results are added"""
for p in self.population:
if p.candidate_replacement == result.configuration:
if self.objective.lt(p.candidate_replacement, p.config):
# candidate replacement was better, replace it!
p.config = p.candidate_replacement
log.info('better point')
p.candidate_replacement = None
class DifferentialEvolutionAlt(DifferentialEvolution):
def __init__(self, cr=0.2, **kwargs):
kwargs['cr'] = cr
super(DifferentialEvolutionAlt, self).__init__(**kwargs)
register(DifferentialEvolution())
register(DifferentialEvolutionAlt())
for i, k in enumerate(params):
if i < self.n_cross or random.random() < self.cr:
# cfg = x1 + use_f*(x2 - x3)
cfg_params[k].set_linear(1.0, x1, use_f, x2, -use_f, x3)
return cfg
def handle_requested_result(self, result):
"""called when new results are added"""
for p in self.population:
if p.candidate_replacement == result.configuration:
if self.objective.lt(p.candidate_replacement, p.config):
# candidate replacement was better, replace it!
p.config = p.candidate_replacement
log.info('better point')
p.candidate_replacement = None
class DifferentialEvolutionAlt(DifferentialEvolution):
def __init__(self, cr=0.2, **kwargs):
kwargs['cr'] = cr
super(DifferentialEvolutionAlt, self).__init__(**kwargs)
register(DifferentialEvolution())
register(DifferentialEvolutionAlt())
register(DifferentialEvolution(population_size=100, cr=0.2,
name='DifferentialEvolution_20_100'))
return population
def select_parameters(self, params):
random.shuffle(params)
ret_list = params[:self.must_mutate_count]
for param in params[self.must_mutate_count:]:
if random.random() < self.mutation_rate:
ret_list.append(param)
return ret_list
def get_default_operator(self, param_type):
return {'op_name': 'op1_randomize', 'args': [], 'kwargs':{}}
register(RandomThreeParentsComposableTechnique(name='ComposableDiffEvolution',
population_size=30))
register_generator(RandomThreeParentsComposableTechnique)
register_generator(GreedyComposableTechnique)
op_map = {}
ComposableEvolutionaryTechnique.add_to_map(op_map,
PermutationParameter,
"op3_cross", xchoice='op3_cross_CX')
ComposableEvolutionaryTechnique.add_to_map(op_map,
"FloatArray",
"op3_cross", strength=0.4)
register(RandomThreeParentsComposableTechnique(name='ComposableDiffEvolutionCX',
operator_map=op_map,
population_size=30))
