def test_model_type(self):
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random',
model_type='GP')
self.assertTrue(isinstance(method.model, GPModel))
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random',
model_type='sparseGP')
self.assertTrue(isinstance(method.model, GPModel))
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random',
model_type='GP_MCMC')
self.assertTrue(isinstance(method.model, GPModel_MCMC))
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random',
model_type='warpedGP')
self.assertTrue(isinstance(method.model, WarpedGPModel))
def test_initial_value(self):
x = np.array([[0, 0, 0, 0, 0]])
y = np.array([[1]])
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random',
X=x,
Y=y)
x_opt, y_opt = method.get_best_point()
self.assertTrue(np.all(x[0] == x_opt))
self.assertEqual(y[0][0], y_opt)
def test_update_subspace(self):
x = np.array([[0, 0, 0, 0, 0]])
y = np.array([[1]])
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random',
X=x,
Y=y)
method.update_subspace()
self.assertEqual(len(method.subspace_idx), self.subspace_dim_size)
for i, idx in enumerate(method.subspace_idx):
self.assertEqual(str(idx), method.subspace_domain[i]['name'])
def test_update(self):
x = np.array([[0, 0, 0, 0, 0]])
y = np.array([[1]])
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random',
X=x,
Y=y,
normalize_Y=False)
method.update()
self.assertEqual(1, len(method.X_inmodel))
self.assertTrue(
np.all(method.X_inmodel[0] == x[0][method.subspace_idx]))
self.assertEqual(y[0][0], method.Y_inmodel)
def test_acquisition_optimizer_type(self):
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random',
acquisition_optimizer_type='lbfgs')
self.assertEqual(method.acquisition_optimizer_type, 'lbfgs')
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random',
acquisition_optimizer_type='DIRECT')
self.assertEqual(method.acquisition_optimizer_type, 'DIRECT')
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random',
acquisition_optimizer_type='CMA')
self.assertEqual(method.acquisition_optimizer_type, 'CMA')
def test_acquisition_type(self):
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random',
acquisition_type='EI')
self.assertTrue(isinstance(method.acquisition, AcquisitionEI))
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random',
acquisition_type='MPI')
self.assertTrue(isinstance(method.acquisition, AcquisitionMPI))
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random',
acquisition_type='LCB')
self.assertTrue(isinstance(method.acquisition, AcquisitionLCB))
def test_domain(self):
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random')
self.assertEqual(method.domain[0]['name'], '0')
self.assertEqual(method.domain[1]['name'], '1')
self.assertEqual(method.domain[2]['name'], '2')
self.assertEqual(method.domain[3]['name'], '3')
self.assertEqual(method.domain[4]['name'], '4')
self.assertEqual(len(self.domain),
method.space.objective_dimensionality)
from bayopt.methods.dropout import Dropout
from bayopt.objective_examples.experiments import SchwefelsFunction
import numpy as np
domain = [{'name': 'x0', 'type': 'continuous', 'domain': (-1, 1), 'dimensionality': 1},
{'name': 'x1', 'type': 'continuous', 'domain': (-1, 1), 'dimensionality': 1},
{'name': 'x2', 'type': 'continuous', 'domain': (-1, 1), 'dimensionality': 1},
{'name': 'x3', 'type': 'continuous', 'domain': (-1, 1), 'dimensionality': 1},
{'name': 'x4', 'type': 'continuous', 'domain': (-1, 1), 'dimensionality': 1},
]
dim = len(domain)
fill_in_strategy = 'random'
f = SchwefelsFunction()
X = np.array([np.full(dim, 1)])
method = Dropout(f=f, domain=domain, subspace_dim_size=2, fill_in_strategy=fill_in_strategy, maximize=False,
X=X)
method.run_optimization(max_iter=300, eps=-1)
{
'name': 'x27',
'type': 'continuous',
'domain': (1, 4),
'dimensionality': 1
},
{
'name': 'x28',
'type': 'continuous',
'domain': (1, 4),
'dimensionality': 1
},
{
'name': 'x29',
'type': 'continuous',
'domain': (1, 4),
'dimensionality': 1
},
]
dim = len(domain)
fill_in_strategy = 'random'
f = GaussianMixtureFunction(dim=dim, mean_1=2, mean_2=3)
X = np.array([np.full(dim, 1)])
method = Dropout(f=f,
domain=domain,
subspace_dim_size=5,
fill_in_strategy=fill_in_strategy,
maximize=True)
method.run_optimization(max_iter=200)
},
{
'name': 'x29',
'type': 'continuous',
'domain': (-1, 1),
'dimensionality': 1
},
]
for i in range(5):
fill_in_strategy = 'random'
f = SchwefelsFunction()
method = Dropout(f=f,
domain=domain,
subspace_dim_size=5,
fill_in_strategy=fill_in_strategy,
maximize=False)
method.run_optimization(max_iter=500, eps=0)
fill_in_strategy = 'copy'
f = SchwefelsFunction()
method = Dropout(
f=f,
domain=domain,
subspace_dim_size=1,
fill_in_strategy=fill_in_strategy,
maximize=False,
)
method.run_optimization(max_iter=500, eps=0)
def test_random_fill_in(self):
method = Dropout(f=self.f,
domain=self.domain,
subspace_dim_size=self.subspace_dim_size,
fill_in_strategy='random')
