def log_prior(self, leaf_values):
return T.mean(
log_normal(leaf_values,
T.zeros_like(leaf_values, dtype='float32'),
T.ones_like(leaf_values, dtype='float32'),
self.embedding_size,
dim=2))
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))
log_q_zx = log_normal(z, q_mu, q_sigma)
p_network = Vector(D, placeholder=z, is_input=False) >> Repeat(Tanh(200), 2)
p_mu_network = p_network >> Linear(X.shape[1])
p_mu = p_mu_network.get_outputs()[0].get_placeholder()
p_sigma_network = p_network >> Linear(X.shape[1])
p_sigma = tf.sqrt(tf.exp(p_sigma_network.get_outputs()[0].get_placeholder()))
log_p_xz = log_normal(x, p_mu, p_sigma)# + tf.reduce_sum(tree_log_prob)
with T.session() as sess:
batch = set(sample_minibatch(N, M))
subtree = tree.induced_subtree(batch)
s = np.array([[a.get_node_id(), b.get_node_id(), int(b.is_leaf())] + b.get_node_stats() for a, b in subtree.get_segments()])
def log_likelihood(self, states, costs):
return T.zeros_like(costs)
def _sample(self, num_samples):
a = self.get_parameters('natural')[Stats.X]
d = self.shape()[-1]
gumbel_noise = Gumbel(T.zeros_like(a), T.ones_like(a)).sample(num_samples)
return T.one_hot(T.argmax(a[None] + gumbel_noise, -1), d)