def calibrate_final_tuning(args, pool, hyperparameters: HyperParameters):
sampling = MixedVariableSampling(mask, {
# Integer numbers are sampled via Uniform Random Sampling
'int': get_sampling('int_random')
})
crossover = MixedVariableCrossover(mask, {
'int': get_crossover('real_sbx', prob=0.9, eta=3.0)
})
mutation = MixedVariableMutation(mask, {
# Integer numbers are mutated via polynomial mutation
'int': get_mutation('int_pm', eta=3.0)
})
problem = MetaProblem(args, hyperparameters=hyperparameters, parallelization=None)
algorithm = NSGA2(
pop_size=EXECUTIONS_FOR_ITERATION,
sampling=sampling,
crossover=crossover,
mutation=mutation,
eliminate_duplicates=True,
)
termination = utils.get_termination_for_final_tuning(time=ITERATIONS_TIME)
res, \
best_solution, \
min_average, \
min_stddev = utils.get_minimizer(problem, algorithm, termination)
save_csv(args, best_solution, min_average, min_stddev)
def calibrate_variables(args, pool) -> HyperParameters:
sampling = MixedVariableSampling(mask, {
# Real numbers are sampled via Latin Hypercube Sampling
'real': get_sampling('real_lhs'),
# Integer numbers are sampled via Uniform Random Sampling
'int': get_sampling('int_random')
})
crossover = MixedVariableCrossover(mask, {
'real': get_crossover('real_sbx', prob=0.9, eta=3.0),
'int': get_crossover('real_sbx', prob=0.9, eta=3.0)
})
mutation = MixedVariableMutation(mask, {
# Real numbers are mutated via polynomial mutation
'real': get_mutation('real_pm', eta=3.0),
# Integer numbers are mutated via polynomial mutation
'int': get_mutation('int_pm', eta=3.0)
})
problem = MetaProblem(args, parallelization=('starmap', pool.starmap))
algorithm = NSGA2(
pop_size=EXECUTIONS_FOR_ITERATION,
sampling=sampling,
crossover=crossover,
mutation=mutation,
eliminate_duplicates=True,
)
termination = utils.get_termination_for_variables(max_iterations=MAX_ITERATIONS)
res, \
best_solution, \
min_average, \
min_stddev = utils.get_minimizer(problem, algorithm, termination)
best_solution = res.X[0]
min_average = res.F[0][0]
min_stddev = res.F[0][1]
mutation_probability = get_var(best_solution, 'mutation_probability')
crossover_rate = get_var(best_solution, 'crossover_rate')
mu = get_var(best_solution, 'mu')
lambda_ = get_var(best_solution, 'lambda')
k = get_var(best_solution, 'k')
hyperparameters = HyperParameters(
mutation_probability=mutation_probability,
crossover_rate=crossover_rate,
mu=mu,
lambda_=lambda_,
k=k
)
save_csv(args, best_solution, min_average, min_stddev)
return hyperparameters
def get_mutation(self):
mask = []
for name in (self.space.numeric_names + self.space.enum_names):
if self.space.paras[name].is_discrete_after_transform:
mask.append('int')
else:
mask.append('real')
mutation = MixedVariableMutation(mask, {
'real' : get_mutation('real_pm', eta = 20),
'int' : get_mutation('int_pm', eta = 20)
})
return mutation
def get_operators(config):
if config.config == "DeepMindBigGAN":
mask = ["real"]*config.dim_z + ["bool"]*config.num_classes
real_sampling = None
if config.config == "DeepMindBigGAN":
real_sampling = TruncatedNormalRandomSampling()
sampling = MixedVariableSampling(mask, {
"real": real_sampling,
"bool": BinaryRandomSampling(prob=5/1000)
})
crossover = MixedVariableCrossover(mask, {
"real": get_crossover("real_sbx", prob=1.0, eta=3.0),
"bool": get_crossover("bin_hux", prob=0.2)
})
mutation = MixedVariableMutation(mask, {
"real": get_mutation("real_pm", prob=0.5, eta=3.0),
"bool": get_mutation("bin_bitflip", prob=10/1000)
})
return dict(
sampling=sampling,
crossover=crossover,
mutation=mutation
)
elif config.config.split("_")[0] == "StyleGAN2":
return dict(
sampling=NormalRandomSampling(),
crossover=get_crossover("real_sbx", prob=1.0, eta=3.0),
mutation=get_mutation("real_pm", prob=0.5, eta=3.0)
)
elif config.config == "GPT2":
return dict(
sampling=get_sampling("int_random"),
crossover=get_crossover("int_sbx", prob=1.0, eta=3.0),
mutation=get_mutation("int_pm", prob=0.5, eta=3.0)
)
else:
raise Exception("Unknown config")
def run_multiJ(dict_t):
# def run_multiJ():
mask = [
"real", "int", "real", "real", "int", "int", "int", "int", "int",
"int", "int"
]
sampling = MixedVariableSampling(mask, {
"real": get_sampling("real_random"),
"int": get_sampling("int_random")
})
crossover = MixedVariableCrossover(
mask, {
"real": get_crossover("real_sbx", prob=1.0, eta=3.0),
"int": get_crossover("int_sbx", prob=1.0, eta=3.0)
})
mutation = MixedVariableMutation(
mask, {
"real": get_mutation("real_pm", eta=3.0),
"int": get_mutation("int_pm", eta=3.0)
})
#[V_gBurn,ng,Tdig,debt_level,V_cng_p,e_priceS,farm1,farm2,farm3,farm4,farm5,farm6,farm7]
problem = BiogasMultiJ(dict_t)
# problem = BiogasMultiJ()
algorithm = NSGA2(
pop_size=dict_t['NSGA_pop'],
sampling=sampling,
crossover=crossover,
n_offsprings=dict_t['NSGA_off'],
mutation=mutation,
eliminate_duplicates=True,
)
res = minimize(problem,
algorithm, ("n_gen", dict_t['NSGA_gen']),
verbose=True,
seed=1,
save_history=True)
return res
def create_attack(
initial_state,
model,
scaler,
encoder,
n_gen,
pop_size,
n_offsprings,
):
problem = VenusProblem(initial_state, model, encoder, scaler)
type_mask = [
"real",
"int",
"real",
"real",
"real",
"real",
"real",
"real",
"real",
"real",
"real",
"int",
"real",
"real",
"int",
]
sampling = MixedVariableSampling(
type_mask,
{
"real": get_sampling("real_random"),
"int": get_sampling("int_random")
},
)
# Default parameters for crossover (prob=0.9, eta=30)
crossover = MixedVariableCrossover(
type_mask,
{
"real": get_crossover("real_sbx", prob=0.9, eta=30),
"int": get_crossover("int_sbx", prob=0.9, eta=30),
},
)
# Default parameters for mutation (eta=20)
mutation = MixedVariableMutation(
type_mask,
{
"real": get_mutation("real_pm", eta=20),
"int": get_mutation("int_pm", eta=20),
},
)
algorithm = NSGA2(
pop_size=pop_size,
n_offsprings=n_offsprings,
sampling=sampling,
crossover=crossover,
mutation=mutation,
eliminate_duplicates=True,
return_least_infeasible=True,
)
termination = get_termination("n_gen", n_gen)
return problem, algorithm, termination
mask = ["int", "int","int","int","int","int","int","int","int","int","int","int"]
from pymoo.factory import get_sampling, get_crossover, get_mutation
from pymoo.operators.mixed_variable_operator import MixedVariableSampling, MixedVariableMutation, MixedVariableCrossover
sampling = MixedVariableSampling(mask, {
"real": get_sampling("real_random"),
"int": get_sampling("int_random")
})
crossover = MixedVariableCrossover(mask, {
"real": get_crossover("real_sbx", prob=1.0, eta=3.0),
"int": get_crossover("int_sbx", prob=1.0, eta=3.0)
})
mutation = MixedVariableMutation(mask, {
"real": get_mutation("real_pm", eta=3.0),
"int": get_mutation("int_pm", eta=3.0)
})
ref_dirs = get_reference_directions("das-dennis", 3, n_partitions=12)
algorithm = NSGA3(ref_dirs)
termination = get_termination("f_tol", tol=0.001, n_last=20, n_max_gen=1000, nth_gen=10)
termination = ("n_gen", 1000)
from pymoo.factory import get_crossover, get_mutation, get_sampling
from pymoo.optimize import minimize
problem = MyProblem()
res = minimize(MyProblem(),
algorithm,
termination,
seed=1,
pf=problem.pareto_front(use_cache=False),