def setUp(self):
self.n = 1
self.x_train = [[1.0]]
self.y_train = [[1.0]]
self.num_latent = 1
self.num_outputs = 1
self.num_weights = 1
self.num_components = 1
self.num_inducing = 1
self.inducing_locations = [[1.0]]
self.kern_f = [
SE(num_dimensions=1, ARD=True, length_scale=1.0, sigma=1.0)
]
self.kern_w = [[
SE(num_dimensions=1, ARD=True, length_scale=1.0, sigma=1.0)
]]
self.q_means_u = [[[1.0]]]
self.q_covars_u = [[[2.0]]]
self.q_means_v = [[[[3.0]]]]
self.q_covars_v = [[[[4.0]]]]
self.q_weights = [1.0]
self.sigma_y = 2.0
self.inference = Inference(tf.constant(self.x_train),
tf.constant(self.y_train), self.num_latent,
self.num_outputs, self.num_weights,
self.num_components, self.num_inducing,
tf.constant(self.inducing_locations),
self.kern_f, self.kern_w,
tf.constant(self.q_means_u),
tf.constant(self.q_covars_u),
tf.constant(self.q_means_v),
tf.constant(self.q_covars_v),
tf.constant(self.q_weights), self.sigma_y)
print("\n In method", self._testMethodName)
def test_kernel_symmetrix(self):
#Arrange
se = SE()
x1 = tf.constant([1.0, 2.0])
x2 = tf.constant([1.0, 2.0])
#Act
init = tf.global_variables_initializer()
with tf.Session() as session:
session.run(init)
v = se.kernel(x1, x2)
#Assert
print(v.eval())
def test_covar_is_sigma_for_same_value__scalar(self):
#Arrange
se = SE(num_dimensions=1, ARD=True, length_scale=1.0, sigma=2.0)
x1 = tf.constant([[1.0]])
x2 = tf.constant([[1.0]])
#Act
init = tf.global_variables_initializer()
with tf.Session() as session:
session.run(init)
v = se.kernel(x1, x2)
sigma = se.get_parameters()['sigma']
#Assert
assert (v.eval() == sigma.eval()**2)
def test_length_scales(self):
#Arrange
ls = 3.0
se = SE(num_dimensions=1, ARD=True, length_scale=ls, sigma=1.0)
x1 = tf.constant([[1.0]])
x2 = tf.constant([[2.0]])
#Act
init = tf.global_variables_initializer()
with tf.Session() as session:
session.run(init)
k = se.kernel(x1, x2)
#Assert
np.testing.assert_almost_equal(k.eval(), np.exp(-1 / (2 * ls**2)))
def test_covar_is_symmetric__scalar(self):
#Arrange
se = SE(num_dimensions=1, ARD=True, length_scale=1.0, sigma=1.0)
x1 = tf.constant([[1.0], [2.0], [3.0]])
x2 = tf.constant([[1.0], [2.0], [3.0]])
#Act
init = tf.global_variables_initializer()
with tf.Session() as session:
session.run(init)
k = se.kernel(x1, x2)
k_lower = tf.matrix_band_part(k, -1, 0)
k_upper = tf.matrix_band_part(k, 0, -1)
k_upper_T = tf.transpose(k_upper)
#Assert
assert (tf.equal(k_lower, k_upper_T).eval().all())
def test_psd__1_d(self):
#Arrange
ls = 3.0
num_dim = 2
se = SE(num_dimensions=num_dim, ARD=False, length_scale=ls, sigma=1.0)
#x1 = tf.constant([[ 0. ], [ 0.12822827], [ 0.25645654], [ 0.38468481], [ 0.51291309], [ 0.64114136], [ 0.76936963], [ 0.8975979 ], [ 1.02582617], [ 1.15405444], [ 1.28228272], [ 1.41051099], [ 1.53873926], [ 1.66696753], [ 1.7951958 ], [ 1.92342407], [ 2.05165235], [ 2.17988062], [ 2.30810889], [ 2.43633716], [ 2.56456543], [ 2.6927937 ], [ 2.82102197], [ 2.94925025], [ 3.07747852], [ 3.20570679], [ 3.33393506], [ 3.46216333], [ 3.5903916 ], [ 3.71861988], [ 3.84684815], [ 3.97507642], [ 4.10330469], [ 4.23153296], [ 4.35976123], [ 4.48798951], [ 4.61621778], [ 4.74444605], [ 4.87267432], [ 5.00090259], [ 5.12913086], [ 5.25735913], [ 5.38558741], [ 5.51381568], [ 5.64204395], [ 5.77027222], [ 5.89850049], [ 6.02672876], [ 6.15495704], [ 6.28318531]])
x1 = tf.constant([[float(i)] for i in range(0, 10)])
x2 = x1
#Act
init = tf.global_variables_initializer()
with tf.Session() as session:
session.run(init)
k = se.kernel(x1, x2, jitter=True)
#Assert
result = tf.self_adjoint_eig(k)[0]
result = result.eval().flatten()
assert (all(i >= 0.0 for i in result))
def test_is_correct__1_d(self):
#Arrange
ls = 1.0
sig = 1.0
num_dim = 1
se = SE(num_dimensions=num_dim, ARD=False, length_scale=ls, sigma=sig)
#x1 = tf.constant([[ 0. ], [ 0.12822827], [ 0.25645654], [ 0.38468481], [ 0.51291309], [ 0.64114136], [ 0.76936963], [ 0.8975979 ], [ 1.02582617], [ 1.15405444], [ 1.28228272], [ 1.41051099], [ 1.53873926], [ 1.66696753], [ 1.7951958 ], [ 1.92342407], [ 2.05165235], [ 2.17988062], [ 2.30810889], [ 2.43633716], [ 2.56456543], [ 2.6927937 ], [ 2.82102197], [ 2.94925025], [ 3.07747852], [ 3.20570679], [ 3.33393506], [ 3.46216333], [ 3.5903916 ], [ 3.71861988], [ 3.84684815], [ 3.97507642], [ 4.10330469], [ 4.23153296], [ 4.35976123], [ 4.48798951], [ 4.61621778], [ 4.74444605], [ 4.87267432], [ 5.00090259], [ 5.12913086], [ 5.25735913], [ 5.38558741], [ 5.51381568], [ 5.64204395], [ 5.77027222], [ 5.89850049], [ 6.02672876], [ 6.15495704], [ 6.28318531]])
x1_val = 2.1
x2_val = 3.2
x1 = tf.constant([[x1_val]])
x2 = tf.constant([[x2_val]])
#Act
init = tf.global_variables_initializer()
with tf.Session() as session:
session.run(init)
k = se.kernel(x1, x2)
#Assert
result = k.eval().flatten()
desired = (sig**2) * np.exp(-((x1_val - x2_val)**2) / (2 *
(ls**2)))
np.testing.assert_almost_equal(result[0], desired)