def get_tf_tensor(self, hyper_parameter: List[tf.Tensor],
x_vector: np.ndarray,
x_vector_: np.ndarray) -> tf.Tensor:
assert x_vector is not None and x_vector_ is not None, "Input vectors x and x_ uninitialized: " + str(
self)
assert len(hyper_parameter) == self.get_number_of_hyper_parameter(
), "Invalid hyper_param size: " + str(self)
with tf.name_scope("Matern_5_2_Kernel"):
dist: tf.Tensor = aux_dist.manhattan_distance(x_vector, x_vector_)
l = tf.abs(hyper_parameter[0])
frac: tf.Tensor = tf.divide(
tf.sqrt(tf.cast(5, dtype=global_param.p_dtype)) * dist, l)
third_summand: tf.Tensor = tf.divide(
tf.cast(5, dtype=global_param.p_dtype) * tf.square(dist),
tf.cast(3, dtype=global_param.p_dtype) * tf.square(l))
result: tf.Tensor = (tf.cast(1, dtype=global_param.p_dtype) +
frac + third_summand) * tf.exp(-frac)
if global_param.p_scaled_base_kernel:
result = tf.multiply(hyper_parameter[1], result)
self.last_hyper_parameter = hyper_parameter
return result
def get_l_derivative_matrix(hyper_parameter: List[tf.Tensor], x_vector: np.ndarray, x_vector_: np.ndarray) \
-> tf.Tensor:
diff: tf.Tensor = aux_dist.manhattan_distance(
x_vector,
x_vector_) #tf.math.subtract(x_vector, tf.transpose(x_vector_))
squared_distance: tf.Tensor = tf.square(diff, name="squared_distance")
length_scale = hyper_parameter[0]
l_squared: tf.Tensor = tf.square(length_scale, name="l_squared")
cov_mat: tf.Tensor = \
tf.math.exp(-0.5 * tf.math.divide(squared_distance, l_squared), name="SquaredExponentialKernel")
return tf.multiply(
tf.divide(diff, tf.multiply(length_scale, l_squared)), cov_mat)
def get_derivative_matrices(self, hyper_parameter: List[tf.Tensor], x_vector: np.ndarray, x_vector_: np.ndarray) -> \
List[tf.Tensor]:
pi = tf.cast(np.pi, global_param.p_dtype)
dist = aux_dist.manhattan_distance(
x_vector, x_vector_
) #tf.math.subtract(tf.reshape(x_vector, [-1, 1]), tf.reshape(x_vector_, [1, -1]))
u = tf.math.multiply(pi, tf.math.divide(dist, hyper_parameter[1]))
sine = tf.square(tf.sin(u))
l_squared = tf.math.square(hyper_parameter[0])
exponent = tf.divide(
tf.multiply(tf.cast(-2, dtype=global_param.p_dtype), sine),
l_squared)
cov_mat = tf.math.exp(exponent, name="PeriodicKernel")
self.last_hyper_parameter = hyper_parameter
self.latest_cov_mat = cov_mat
return [
self.get_l_derivative_matrix(hyper_parameter, cov_mat, exponent),
self.get_p_derivative_matrix(hyper_parameter, cov_mat, pi, dist, u,
l_squared)
]
def get_tf_tensor(self, hyper_parameter: List[tf.Tensor],
x_vector: np.ndarray,
x_vector_: np.ndarray) -> tf.Tensor:
assert x_vector is not None and x_vector_ is not None, "Input vectors x and x_ uninitialized: " + str(
self)
assert len(hyper_parameter) == self.get_number_of_hyper_parameter(
), "Invalid hyper_param size: " + str(self)
with tf.name_scope("PeriodicKernel"):
dist: tf.Tensor = aux_dist.manhattan_distance(x_vector, x_vector_)
sine_input: tf.Tensor = tf.math.multiply(
tf.cast(np.pi, global_param.p_dtype),
tf.math.divide(dist, hyper_parameter[1]))
sine: tf.Tensor = tf.square(tf.math.sin(sine_input))
result: tf.Tensor = tf.math.exp(-2 * sine /
tf.math.square(hyper_parameter[0]),
name="PeriodicKernel")
if global_param.p_scaled_base_kernel:
result = tf.multiply(hyper_parameter[2], result)
self.last_hyper_parameter = hyper_parameter
return result
def get_c_derivative_matrix(hyper_parameter: List[tf.Tensor], x_vector: np.ndarray, x_vector_: np.ndarray) \
-> tf.Tensor:
diff = aux_dist.manhattan_distance(x_vector, x_vector_)
return tf.add(
diff,
tf.multiply(tf.cast(2, global_param.p_dtype), hyper_parameter[0]))