def test_randomize_hyperparameters_samples_different_values_for_multi_dimensional_params(
) -> None:
kernel = gpflow.kernels.RBF(variance=1.0, lengthscales=[0.2, 0.2])
upper = tf.cast([10.0] * 2, dtype=tf.float64)
lower = upper / 100
kernel.lengthscales = gpflow.Parameter(kernel.lengthscales,
transform=tfp.bijectors.Sigmoid(
low=lower, high=upper))
randomize_hyperparameters(kernel)
npt.assert_raises(AssertionError, npt.assert_allclose,
kernel.lengthscales[0], kernel.lengthscales[1])
def test_randomize_hyperparameters_randomizes_kernel_parameters_with_priors(
dim: int) -> None:
kernel = gpflow.kernels.RBF(variance=1.0, lengthscales=[0.2] * dim)
kernel.lengthscales.prior = tfp.distributions.LogNormal(loc=tf.math.log(
kernel.lengthscales),
scale=1.0)
randomize_hyperparameters(kernel)
npt.assert_allclose(1.0, kernel.variance)
npt.assert_array_equal(dim, kernel.lengthscales.shape)
npt.assert_raises(AssertionError, npt.assert_allclose, [0.2] * dim,
kernel.lengthscales)
def test_randomize_hyperparameters_randomizes_constrained_kernel_parameters(
dim: int) -> None:
kernel = gpflow.kernels.RBF(variance=1.0, lengthscales=[0.2] * dim)
upper = tf.cast([10.0] * dim, dtype=tf.float64)
lower = upper / 100
kernel.lengthscales = gpflow.Parameter(kernel.lengthscales,
transform=tfp.bijectors.Sigmoid(
low=lower, high=upper))
randomize_hyperparameters(kernel)
npt.assert_allclose(1.0, kernel.variance)
npt.assert_array_equal(dim, kernel.lengthscales.shape)
npt.assert_raises(AssertionError, npt.assert_allclose, [0.2] * dim,
kernel.lengthscales)
def test_randomize_hyperparameters_randomizes_kernel_parameters_with_constraints_or_priors(
dim: int, ) -> None:
kernel = gpflow.kernels.RBF(variance=1.0, lengthscales=[0.2] * dim)
upper = tf.cast([10.0] * dim, dtype=tf.float64)
lower = upper / 100
kernel.lengthscales = gpflow.Parameter(kernel.lengthscales,
transform=tfp.bijectors.Sigmoid(
low=lower, high=upper))
kernel.variance.prior = tfp.distributions.LogNormal(loc=np.float64(-2.0),
scale=np.float64(1.0))
randomize_hyperparameters(kernel)
npt.assert_raises(AssertionError, npt.assert_allclose, 1.0,
kernel.variance)
npt.assert_array_equal(dim, kernel.lengthscales.shape)
npt.assert_raises(AssertionError, npt.assert_allclose, [0.2] * dim,
kernel.lengthscales)
def test_randomize_hyperparameters_samples_from_constraints_when_given_prior_and_constraint(
dim: int, ) -> None:
kernel = gpflow.kernels.RBF(variance=1.0, lengthscales=[0.2] * dim)
upper = tf.cast([0.5] * dim, dtype=tf.float64)
lower = upper / 100
kernel.lengthscales = gpflow.Parameter(kernel.lengthscales,
transform=tfp.bijectors.Sigmoid(
low=lower, high=upper))
kernel.lengthscales.prior = tfp.distributions.Uniform(low=10.0, high=100.0)
kernel.variance.prior = tfp.distributions.LogNormal(loc=np.float64(-2.0),
scale=np.float64(1.0))
randomize_hyperparameters(kernel)
npt.assert_array_less(kernel.lengthscales, [0.5] * dim)
npt.assert_raises(AssertionError, npt.assert_allclose, [0.2] * dim,
kernel.lengthscales)