def _build_gaussian_process_test_data(self, test_environment):
"""Build up a Gaussian Process randomly by repeatedly drawing from and then adding to the prior.
:param test_environment: parameters describing how to construct a GP prior
:type test_environment: GaussianProcessTestEnvironmentInput
:return: gaussian process environments that can be used to run tests
:rtype: GaussianProcessTestEnvironment
"""
covariance = gp_utils.fill_random_covariance_hyperparameters(
test_environment.hyperparameter_interval,
test_environment.num_hyperparameters,
covariance_type=test_environment.covariance_class,
)
domain_bounds = gp_utils.fill_random_domain_bounds(
test_environment.lower_bound_interval,
test_environment.upper_bound_interval,
test_environment.dim,
)
domain = test_environment.spatial_domain_class(ClosedInterval.build_closed_intervals_from_list(domain_bounds))
points_sampled = domain.generate_uniform_random_points_in_domain(test_environment.num_sampled)
gaussian_process = gp_utils.build_random_gaussian_process(
points_sampled,
covariance,
noise_variance=test_environment.noise_variance,
gaussian_process_type=test_environment.gaussian_process_class,
)
return GaussianProcessTestEnvironment(domain, gaussian_process)
def _build_gaussian_process_test_data(self, test_environment):
"""Build up a Gaussian Process randomly by repeatedly drawing from and then adding to the prior.
:param test_environment: parameters describing how to construct a GP prior
:type test_environment: GaussianProcessTestEnvironmentInput
:return: gaussian process environments that can be used to run tests
:rtype: GaussianProcessTestEnvironment
"""
covariance = gp_utils.fill_random_covariance_hyperparameters(
test_environment.hyperparameter_interval,
test_environment.num_hyperparameters,
covariance_type=test_environment.covariance_class,
)
domain_bounds = gp_utils.fill_random_domain_bounds(
test_environment.lower_bound_interval,
test_environment.upper_bound_interval,
test_environment.dim,
)
domain = test_environment.spatial_domain_class(
ClosedInterval.build_closed_intervals_from_list(domain_bounds))
points_sampled = domain.generate_uniform_random_points_in_domain(
test_environment.num_sampled)
gaussian_process = gp_utils.build_random_gaussian_process(
points_sampled,
covariance,
noise_variance=test_environment.noise_variance,
gaussian_process_type=test_environment.gaussian_process_class,
)
return GaussianProcessTestEnvironment(domain, gaussian_process)
def test_hyperparameter_gradient_pings(self):
"""Ping test (compare analytic result to finite difference) the gradient wrt hyperparameters."""
h = 2.0e-3
tolerance = 4.0e-5
num_tests = 10
dim = 3
num_hyperparameters = dim + 1
hyperparameter_interval = ClosedInterval(3.0, 5.0)
domain = TensorProductDomain(
ClosedInterval.build_closed_intervals_from_list([[-1.0, 1.0],
[-1.0, 1.0],
[-1.0, 1.0]]))
points1 = domain.generate_uniform_random_points_in_domain(num_tests)
points2 = domain.generate_uniform_random_points_in_domain(num_tests)
for i in xrange(num_tests):
point_one = points1[i, ...]
point_two = points2[i, ...]
covariance = gp_utils.fill_random_covariance_hyperparameters(
hyperparameter_interval,
num_hyperparameters,
covariance_type=self.CovarianceClass,
)
analytic_grad = covariance.hyperparameter_grad_covariance(
point_one, point_two)
for k in xrange(covariance.num_hyperparameters):
hyperparameters_old = covariance.hyperparameters
# hyperparamter + h
hyperparameters_p = numpy.copy(hyperparameters_old)
hyperparameters_p[k] += h
covariance.hyperparameters = hyperparameters_p
cov_p = covariance.covariance(point_one, point_two)
covariance.hyperparameters = hyperparameters_old
# hyperparamter - h
hyperparameters_m = numpy.copy(hyperparameters_old)
hyperparameters_m[k] -= h
covariance.hyperparameters = hyperparameters_m
cov_m = covariance.covariance(point_one, point_two)
covariance.hyperparameters = hyperparameters_old
# calculate finite diff
fd_grad = (cov_p - cov_m) / (2.0 * h)
self.assert_scalar_within_relative(fd_grad, analytic_grad[k],
tolerance)
def test_hyperparameter_gradient_pings(self):
"""Ping test (compare analytic result to finite difference) the gradient wrt hyperparameters."""
h = 2.0e-3
tolerance = 4.0e-5
num_tests = 10
dim = 3
num_hyperparameters = dim + 1
hyperparameter_interval = ClosedInterval(3.0, 5.0)
domain = TensorProductDomain(ClosedInterval.build_closed_intervals_from_list([[-1.0, 1.0], [-1.0, 1.0], [-1.0, 1.0]]))
points1 = domain.generate_uniform_random_points_in_domain(num_tests)
points2 = domain.generate_uniform_random_points_in_domain(num_tests)
for i in xrange(num_tests):
point_one = points1[i, ...]
point_two = points2[i, ...]
covariance = gp_utils.fill_random_covariance_hyperparameters(
hyperparameter_interval,
num_hyperparameters,
covariance_type=self.CovarianceClass,
)
analytic_grad = covariance.hyperparameter_grad_covariance(point_one, point_two)
for k in xrange(covariance.num_hyperparameters):
hyperparameters_old = covariance.hyperparameters
# hyperparamter + h
hyperparameters_p = numpy.copy(hyperparameters_old)
hyperparameters_p[k] += h
covariance.hyperparameters = hyperparameters_p
cov_p = covariance.covariance(point_one, point_two)
covariance.hyperparameters = hyperparameters_old
# hyperparamter - h
hyperparameters_m = numpy.copy(hyperparameters_old)
hyperparameters_m[k] -= h
covariance.hyperparameters = hyperparameters_m
cov_m = covariance.covariance(point_one, point_two)
covariance.hyperparameters = hyperparameters_old
# calculate finite diff
fd_grad = (cov_p - cov_m) / (2.0 * h)
self.assert_scalar_within_relative(fd_grad, analytic_grad[k], tolerance)
