def initialize_objective(self):
self.C, self.c = (
T.variable(T.random_normal([self.ds, self.ds])),
T.variable(T.random_normal([self.ds])),
)
if self.learn_stdev:
self.stdev = T.variable(T.to_float(self.cost_stdev))
else:
self.stdev = T.to_float(self.cost_stdev)
def initialize_objective(self):
H, ds, da = self.horizon, self.ds, self.da
if self.time_varying:
A = T.concatenate([T.eye(ds), T.zeros([ds, da])], -1)
self.A = T.variable(A[None] + 1e-2 * T.random_normal([H - 1, ds, ds + da]))
self.Q_log_diag = T.variable(T.random_normal([H - 1, ds]) + 1)
self.Q = T.matrix_diag(T.exp(self.Q_log_diag))
else:
A = T.concatenate([T.eye(ds), T.zeros([ds, da])], -1)
self.A = T.variable(A + 1e-2 * T.random_normal([ds, ds + da]))
self.Q_log_diag = T.variable(T.random_normal([ds]) + 1)
self.Q = T.matrix_diag(T.exp(self.Q_log_diag))
def create_parameter(self, name, shape, initial_value=None):
if name not in self.parameters:
if initial_value is None:
parameter = T.variable(
initialize_weights(self.initialization, shape),
name=name,
)
else:
parameter = T.variable(
np.array(initial_value, dtype=T.floatx()),
name=name,
)
self.parameters[name] = parameter
def orthogonal(shape, scale=1.1):
flat_shape = (shape[0], np.prod(shape[1:]))
a = np.random.normal(0.0, 1.0, flat_shape)
u, _, v = np.linalg.svd(a, full_matrices=False)
# Pick the one with the correct shape.
q = u if u.shape == flat_shape else v
q = q.reshape(shape)
return T.variable(scale * q[:shape[0], :shape[1]])
a_idx = segments[:, 0]
b_idx = segments[:, 1]
leaf_segment = segments[:, 2]
m = segments[:, 3]
log_fac = segments[:, 4]
x = T.matrix(name='x')
e = T.matrix(name='e')
q_network = Vector(X.shape[1], placeholder=x, is_input=False) >> Repeat(Tanh(200), 2)
q_mu_network = q_network >> Linear(D)
q_mu = q_mu_network.get_outputs()[0].get_placeholder()
q_sigma_network = q_network >> Linear(D)
q_sigma = tf.sqrt(tf.exp(q_sigma_network.get_outputs()[0].get_placeholder()))
z = q_mu + e * q_sigma
values, times = T.variable(values), T.variable(times)
values = tf.concat(0, [z, values])
harmonic = T.variable(create_harmonic(M))
a_batch_values = T.gather(values, a_idx)
a_batch_times = T.gather(times, a_idx)
b_batch_values = T.gather(values, b_idx)
b_batch_times = T.gather(times, b_idx)
harmonic_m = T.gather(harmonic, m - 1)
time_delta = b_batch_times - a_batch_times
normal_log_prob = log_normal(b_batch_times, a_batch_times, time_delta * 1.0 / 1)
log_pt = (log_a(a_batch_times) + (A(a_batch_times) - A(b_batch_times)) * harmonic_m)
tree_log_prob = tf.select(tf.cast(leaf_segment, 'bool'), T.zeros_like(a_batch_times), log_pt + tf.to_float(log_fac))
def make_variable(dist):
return dist.__class__(T.variable(dist.get_parameters('natural')),
parameter_type='natural')
def identity(shape, scale=1):
if len(shape) != 2 or shape[0] != shape[1]:
raise ValueError('Identity matrix initialization can only be used '
'for 2D square matrices.')
else:
return T.variable(scale * np.identity(shape[0]))
def as_variable(self):
params = self.get_parameters('natural')
return self.__class__(
{stat: T.variable(params[stat])
for stat in self.statistics()}, 'natural')