def test_non_linear_kernel_ard():
"""
Test that the kernels that act on the input space have the correct number of lengthscales when ARD is true
"""
kernels = make_non_linear_kernels(GPy.kern.RBF, 2, 2, ARD=True)
assert len(kernels[0].lengthscale) == 2
assert len(kernels[1].bias_kernel_fidelity2.lengthscale) == 2
assert len(kernels[1].mul.scale_kernel_fidelity2.lengthscale) == 2
def test_non_linear_kernel_ard():
"""
Test that the kernels that act on the input space have the correct number of lengthscales when ARD is true
"""
kernels = make_non_linear_kernels(GPy.kern.RBF, 2, 2, ARD=True)
assert len(kernels[0].lengthscale) == 2
assert len(kernels[1].bias_kernel_fidelity2.lengthscale) == 2
assert len(kernels[1].mul.scale_kernel_fidelity2.lengthscale) == 2
def non_linear_model(self, x_init, y_init):
"""
Creates a NonLinearModel instance to use in tests
"""
np.random.seed(123)
base_kernel = GPy.kern.RBF
kernel = make_non_linear_kernels(base_kernel, len(x_init), x_init.shape[1] - 1)
model = emukit.multi_fidelity.models.NonLinearMultiFidelityModel(x_init, y_init, 3, kernel, n_samples=3)
return model
def non_linear_model(self, x_init, y_init):
"""
Creates a NonLinearModel instance to use in tests
"""
np.random.seed(123)
base_kernel = GPy.kern.RBF
kernel = make_non_linear_kernels(base_kernel, len(x_init),
x_init.shape[1] - 1)
model = emukit.multi_fidelity.models.NonLinearMultiFidelityModel(
x_init, y_init, 3, kernel, n_samples=3)
return model
def log_nonlinear_mfdgp(X_train, Y_train):
base_kernel = GPy.kern.RBF
kernels = make_non_linear_kernels(base_kernel, 2, X_train.shape[1] - 1)
nonlin_mf_model = NonLinearMultiFidelityModel(
X_train, Y_train, n_fidelities=2, kernels=kernels, verbose=True,
optimization_restarts=5)
mf1, mf2 = nonlin_mf_model.models
mf1.Gaussian_noise.variance.fix(0)
mf2.Gaussian_noise.variance.fix(0)
nonlin_mf_model.optimize()
return nonlin_mf_model
def variable_model(x_train, y_train, n_fidelity, fullname_temp):
base_kernel = GPy.kern.RBF
kernels = make_non_linear_kernels(base_kernel, 2, x_train.shape[1] - 1)
nonlin_mf_model = NonLinearMultiFidelityModel(x_train,
y_train,
n_fidelities=n_fidelity,
kernels=kernels,
verbose=True,
optimization_restarts=5)
for m in nonlin_mf_model.models:
m.Gaussian_noise.variance.fix(0)
nonlin_mf_model.optimize()
# 保存模型
from sklearn.externals import joblib
model_temp = fullname_temp + "\\nonlin_nf_model.pkl"
gol.set_map("model_temp", model_temp)
joblib.dump(nonlin_mf_model, model_temp)
def nonlinear_model(X_train, Y_train, save=False):
""" Returns GPy model of nonlinear multi-fidelity GP (Perdikaris 2017) """
base_kernel = GPy.kern.RBF
kernels = make_non_linear_kernels(base_kernel, 2, X_train.shape[1] - 1)
nonlin_mf_model = NonLinearMultiFidelityModel(X_train,
Y_train,
n_fidelities=2,
kernels=kernels,
verbose=True,
optimization_restarts=5)
for m in nonlin_mf_model.models:
m.Gaussian_noise.variance.fix(0)
# Optimise
nonlin_mf_model.optimize()
if save is not False:
nonlin_mf_model.save_model(save)
return nonlin_mf_model
alpha=0.3)
plt.xlim(0, 1)
plt.xlabel('x')
plt.ylabel('f (x)')
plt.legend(['True Function', 'Linear multi-fidelity GP'], loc='lower right')
plt.title('Linear multi-fidelity model fit to high fidelity function')
####
## Create nonlinear model
from emukit.multi_fidelity.models.non_linear_multi_fidelity_model import make_non_linear_kernels, \
NonLinearMultiFidelityModel
base_kernel = GPy.kern.RBF
kernels = make_non_linear_kernels(base_kernel, 2, X_train.shape[1] - 1)
# print(kernels)
nonlin_mf_model = NonLinearMultiFidelityModel(
X_train,
Y_train,
n_fidelities=2,
kernels=kernels, #n_samples=1,
verbose=False,
optimization_restarts=5)
# print(nonlin_mf_model.models[0])
# print(nonlin_mf_model.models[1])
for m in nonlin_mf_model.models:
m.Gaussian_noise.variance.fix(0)
nonlin_mf_model.optimize()
print()
