def kernel(self, points1, points2=None):
if points2 is None:
points2 = points1
white_noise = self.white * util.eye(tf.shape(points1)[0])
else:
white_noise = 0.0
#This has to be replaced. Is used so as to have a tensor for points1 and points2
points1 = points1 / self.lengthscale * self.lengthscale
points2 = points2 / self.lengthscale * self.lengthscale
magnitude_square1 = tf.expand_dims(tf.reduce_sum(points1**2, 1), 1)
magnitude_square2 = tf.expand_dims(tf.reduce_sum(points2**2, 1), 1)
distances = (magnitude_square1 -
2 * tf.matmul(points1, tf.transpose(points2)) +
tf.transpose(magnitude_square2))
distances_root = tf.sqrt(
distances +
0.05) / self.lengthscale #Numerical stability problem!!
distances_root = tf.clip_by_value(distances_root, 0.0, self.MAX_DIST)
constant = np.sqrt(5.0)
first_term = (1 + constant * distances_root +
5.0 / 3.0 * distances_root**2) * self.std_dev
second_term = tf.exp(-constant * distances_root)
kernel_matrix = tf.multiply(first_term, second_term)
return kernel_matrix + white_noise
def kernel(self, points1, points2=None):
if points2 is None:
points2 = points1
white_noise = self.white * util.eye(tf.shape(points1)[0])
else:
white_noise = 0.0
kern_matrix = np.zeros((points1.shape[0], points2.shape[0]))
#tf.zeros((tf.shape(points1)[0],tf.shape(points2)[0]))
output_list = []
for i in range(kern_matrix.shape[0]):
for j in range(kern_matrix.shape[1]):
distances = util.euclidean2(points1[i, :], points2[j, :])
distances = tf.clip_by_value(distances, 0.0, self.MAX_DIST)
#V_this is to limit the distances between these two values
first_term = tf.scalar_mul(
self.std_dev, 1.0 + tf.scalar_mul(
math.sqrt(3.0) / self.lengthscale, distances))
second_term = tf.exp(
tf.scalar_mul(-math.sqrt(3.0) / self.lengthscale,
distances))
output_list.append(tf.multiply(first_term, second_term))
kernel_matrix = tf.transpose(
tf.reshape((tf.stack(output_list)),
[points2.shape[0], points1.shape[0]]))
return kernel_matrix + white_noise
def kernel(self, points1, points2=None):
if points2 is None:
points2 = points1
white_noise = self.white * util.eye(tf.shape(points1)[0])
else:
white_noise = 0.0
kern = self.recursive_kernel(points1 / self.lengthscale, points2 / self.lengthscale, self.depth)
return (self.std_dev ** 2) * kern + white_noise
def kernel(self, points1, points2=None):
if points2 is None:
points2 = points1
white_noise = self.white * util.eye(tf.shape(points1)[0])
else:
white_noise = 0.0
#This has to be replaced. Is used so as to have a tensor for points1 and points2
points1 = points1 / self.std_dev * self.std_dev
points2 = points2 / self.std_dev * self.std_dev
variance = self.std_dev**2
kernel_matrix = tf.matmul(points1 * variance, tf.transpose(points2))
return kernel_matrix + white_noise
def kernel(self, points1, points2=None):
if points2 is None:
points2 = points1
white_noise = self.white * util.eye(tf.shape(points1)[0])
else:
white_noise = 0.0
points1 = points1 / self.lengthscale
points2 = points2 / self.lengthscale
magnitude_square1 = tf.expand_dims(tf.reduce_sum(points1**2, 1), 1)
magnitude_square2 = tf.expand_dims(tf.reduce_sum(points2**2, 1), 1)
distances = (magnitude_square1 -
2 * tf.matmul(points1, tf.transpose(points2)) +
tf.transpose(magnitude_square2))
distances = tf.clip_by_value(distances, 0.0, self.MAX_DIST)
kern = ((self.std_dev**2) * tf.exp(-distances / 2.0))
return kern + white_noise
def kernel(
self,
points1,
points2=None
): #V_this is just computing the kernel for two points (two vectors), not for N
#points. It gives back one kernel value. Viene fatto per una matrice in generale. Non per una sola osservazione.
if points2 is None:
points2 = points1
white_noise = self.white * util.eye(tf.shape(points1)[0])
else:
white_noise = 0.0
points1 = points1 / self.lengthscale
points2 = points2 / self.lengthscale
magnitude_square1 = tf.expand_dims(tf.reduce_sum(points1**2, 1), 1)
magnitude_square2 = tf.expand_dims(tf.reduce_sum(points2**2, 1), 1)
distances = (magnitude_square1 -
2 * tf.matmul(points1, tf.transpose(points2)) +
tf.transpose(magnitude_square2))
#print("prima di clip", distances)
distances = tf.clip_by_value(distances, 0.0, self.MAX_DIST)
#print("dopo clip", distances)
kern = ((self.std_dev**2) * tf.exp(-distances / 2.0))
return kern + white_noise
