def qz_xy(x, y, n_z, n_particles):
with zs.BayesianNet() as variational:
lz_xy = tf.layers.dense(tf.to_float(tf.concat([x, y], 1)), 500,
activation=tf.nn.relu)
lz_xy = tf.layers.dense(lz_xy, 500, activation=tf.nn.relu)
lz_mean = tf.layers.dense(lz_xy, n_z)
lz_logstd = tf.layers.dense(lz_xy, n_z)
z = zs.Normal('z', lz_mean, logstd=lz_logstd, n_samples=n_particles,
group_ndims=1)
return variational
def labeled_proposal(x, y, z_dim, n_particles):
bn = zs.BayesianNet()
z_mean, z_logstd = qz_xy(x, y, z_dim)
bn.normal("z",
z_mean,
logstd=z_logstd,
n_samples=n_particles,
group_ndims=1,
is_reparameterized=False)
return bn
def q_net(observed, x_dim, z_dim, n_z_per_x):
with zs.BayesianNet(observed=observed) as variational:
x = zs.Empirical('x', tf.int32, (None, x_dim))
lz_x = tf.layers.dense(tf.to_float(x), 500, activation=tf.nn.relu)
lz_x = tf.layers.dense(lz_x, 500, activation=tf.nn.relu)
z_mean = tf.layers.dense(lz_x, z_dim)
z_logstd = tf.layers.dense(lz_x, z_dim)
z = zs.Normal('z', z_mean, logstd=z_logstd, group_ndims=1,
n_samples=n_z_per_x)
return variational
def build_q_net(x, y, z_dim, y_dim):
bn = zs.BayesianNet()
# concatenate x and y
x = tf.concat(axis=1, values=[x, y])
h = tf.layers.dense(x, 500, activation=tf.nn.relu)
h = tf.layers.dense(h, 500, activation=tf.nn.relu)
z_mean = tf.layers.dense(h, z_dim)
z_logstd = tf.layers.dense(h, z_dim)
bn.normal("z", z_mean, logstd=z_logstd, group_ndims=1)
return bn
def labeled_proposal(x, y, n_z, n_particles):
with zs.BayesianNet() as proposal:
z_mean, z_logstd = qz_xy(x, y, n_z)
z = zs.Normal('z',
z_mean,
logstd=z_logstd,
n_samples=n_particles,
group_event_ndims=1,
is_reparameterized=False)
return proposal
def test_session_run_issue_49(self):
# test fix for the bug at https://github.com/thu-ml/zhusuan/issues/49
with zs.BayesianNet(observed={}) as model:
x_mean = tf.zeros([1, 2])
x_logstd = tf.zeros([1, 2])
x = zs.Normal('x', mean=x_mean, logstd=x_logstd, group_ndims=1)
with self.test_session(use_gpu=True) as sess:
sess.run(tf.global_variables_initializer())
_ = sess.run(x)
def vae(observed, x_dim, z_dim, n, n_particles=1):
with zs.BayesianNet(observed=observed) as model:
z_mean = tf.zeros([n, z_dim])
z = zs.Normal('z', z_mean, std=1., group_ndims=1)
lx_z = tf.layers.dense(z, 500, activation=tf.nn.relu)
lx_z = tf.layers.dense(lx_z, 500, activation=tf.nn.relu)
x_logits = tf.layers.dense(lx_z, x_dim)
x_mean = zs.Implicit("x_mean", tf.sigmoid(x_logits), group_ndims=1)
x = zs.Bernoulli('x', x_logits, group_ndims=1, n_samples=n_particles)
return model
def lntm(observed, D, K, V, eta_mean, eta_logstd):
with zs.BayesianNet(observed=observed) as model:
eta = zs.Normal('eta',
tf.tile(tf.expand_dims(eta_mean, 0), [D, 1]),
logstd=tf.tile(tf.expand_dims(eta_logstd, 0), [D, 1]),
group_event_ndims=1)
beta = zs.Normal('beta', tf.zeros([K, V]),
logstd=tf.ones([K, V]) * log_delta,
group_event_ndims=1)
return model
def build_gen(x_dim, z_dim, n, n_particles=1):
bn = zs.BayesianNet()
z_mean = tf.zeros([n, z_dim])
z = bn.normal("z", z_mean, std=1., group_ndims=1, n_samples=n_particles)
h = tf.layers.dense(z, 500, activation=tf.nn.relu)
h = tf.layers.dense(h, 500, activation=tf.nn.relu)
x_logits = tf.layers.dense(h, x_dim)
bn.deterministic("x_mean", tf.sigmoid(x_logits))
bn.bernoulli("x", x_logits, group_ndims=1)
return bn
def build_q_net(x, y, z_dim, n_z_per_x):
bn = zs.BayesianNet()
print(x.shape, y.shape)
cat_x_y = tf.concat([x, y], axis=1)
h = tf.layers.dense(tf.cast(cat_x_y, tf.float32), 500, activation=tf.nn.relu)
h = tf.layers.dense(h, 500, activation=tf.nn.relu)
z_mean = tf.layers.dense(h, z_dim)
z_logstd = tf.layers.dense(h, z_dim)
bn.normal("z", z_mean, logstd=z_logstd, group_ndims=1, n_samples=n_z_per_x)
return bn
def build_gen(y, x_dim, z_dim, n):
bn = zs.BayesianNet()
z = bn.normal("z", tf.zeros([n, z_dim]), std=1., group_ndims=1)
input = tf.concat([z, y], axis=1)
h = tf.layers.dense(input, 500, activation=tf.nn.relu)
h = tf.layers.dense(h, 500, activation=tf.nn.relu)
x_logits = tf.layers.dense(h, x_dim)
x_mean = bn.deterministic("x_mean", tf.sigmoid(x_logits))
x = bn.bernoulli("x", x_logits, group_ndims=1, dtype=tf.float32)
return bn
def vae(observed, n, n_x, n_z):
with zs.BayesianNet(observed=observed) as model:
z_mean = tf.zeros([n, n_z])
z_logstd = tf.zeros([n, n_z])
z = zs.Normal('z', z_mean, logstd=z_logstd, group_event_ndims=1)
lx_z = layers.fully_connected(z, 500)
lx_z = layers.fully_connected(lx_z, 500)
x_logits = layers.fully_connected(lx_z, n_x, activation_fn=None)
x = zs.Bernoulli('x', x_logits, group_event_ndims=1)
return model, x_logits
def mean_field_variational(n_particles):
with zs.BayesianNet() as variational:
z_mean, z_logstd = [], []
for i in range(2):
z_mean.append(tf.Variable(-2.))
z_logstd.append(tf.Variable(-5.))
_ = zs.Normal('z' + str(i + 1),
z_mean[i],
logstd=z_logstd[i],
n_samples=n_particles)
return variational, z_mean, z_logstd
def p_Y_Xw(observed, X, drop_rate, n_basis, net_sizes, n_samples, task):
with zs.BayesianNet(observed=observed) as model:
f = tf.expand_dims(X, 1)
for i in range(len(net_sizes)-1):
f = tf.layers.dense(f, net_sizes[i+1])
if(i < len(net_sizes)-2):
f = tf.nn.relu(f)
f = tf.squeeze(f, [1])
if(task == "classification"):
f = tf.nn.softmax(f)
return model, f, None
def q_net(x, y, z_dim):
with zs.BayesianNet() as variational:
x_plus_y = tf.concat([x, y], axis=1)
lz_x = tf.layers.dense(tf.to_float(x_plus_y),
500,
activation=tf.nn.relu)
lz_x = tf.layers.dense(lz_x, 500, activation=tf.nn.relu)
z_mean = tf.layers.dense(lz_x, z_dim)
z_logstd = tf.layers.dense(lz_x, z_dim)
z = zs.Normal('z', z_mean, logstd=z_logstd, group_ndims=1)
return variational
def vae(observed, x_dim, z_dim, n, y):
with zs.BayesianNet(observed=observed) as model:
z_mean = tf.zeros([n, z_dim])
z = zs.Normal('z', z_mean, std=1., group_ndims=1)
z_plus_y = tf.concat([z, tf.cast(y, tf.float32)], axis=1)
lx_z = tf.layers.dense(z_plus_y, 500, activation=tf.nn.relu)
lx_z = tf.layers.dense(lx_z, 500, activation=tf.nn.relu)
x_logits = tf.layers.dense(lx_z, x_dim)
x_mean = zs.Implicit("x_mean", tf.sigmoid(x_logits), group_ndims=1)
x = zs.Bernoulli('x', x_logits, group_ndims=1)
return model, x_mean
def q_net(x, z_dim):
'''
Encoder: q(z|x)
'''
with zs.BayesianNet() as variational:
lz_x = layers.fully_connected(tf.to_float(x), 500)
lz_x = layers.fully_connected(lz_x, 500)
z_mean = layers.fully_connected(lz_x, z_dim, activation_fn=None)
z_logstd = layers.fully_connected(lz_x, z_dim, activation_fn=None)
z = zs.Normal('z', z_mean, logstd=z_logstd, group_ndims=1)
return variational
def q_net(x, n_h, n_particles):
with zs.BayesianNet() as proposal:
h1_logits = layers.fully_connected(tf.to_float(x), n_h,
activation_fn=None)
h1 = zs.Bernoulli('h1', h1_logits, n_samples=n_particles,
group_ndims=1, dtype=tf.float32)
h2_logits = layers.fully_connected(h1, n_h, activation_fn=None)
h2 = zs.Bernoulli('h2', h2_logits, group_ndims=1, dtype=tf.float32)
h3_logits = layers.fully_connected(h2, n_h, activation_fn=None)
h3 = zs.Bernoulli('h3', h3_logits, group_ndims=1, dtype=tf.float32)
return proposal
def build_gen(x_dim, z_dim, y, n, n_particles=1):
bn = zs.BayesianNet()
z_mean = tf.zeros([n, z_dim])
z = bn.normal("z", z_mean, std=1., group_ndims=1, n_samples=n_particles)
y = tf.reshape(y, [1, tf.shape(y)[0], 10])
y = tf.tile(y, [n_particles, 1, 1])
h = tf.layers.dense(tf.concat([y, z], axis=2), 500, activation=tf.nn.relu)
h = tf.layers.dense(h, 500, activation=tf.nn.relu)
x_logits = tf.layers.dense(h, x_dim)
bn.deterministic("x_mean", tf.sigmoid(x_logits))
bn.bernoulli("x", x_logits, group_ndims=1)
return bn
def q_net(x, z_dim, n_particles):
with zs.BayesianNet() as variational:
lz_x = layers.fully_connected(tf.to_float(x), 500)
lz_x = layers.fully_connected(lz_x, 500)
z_mean = layers.fully_connected(lz_x, z_dim, activation_fn=None)
z_logstd = layers.fully_connected(lz_x, z_dim, activation_fn=None)
z = zs.Normal('z',
z_mean,
logstd=z_logstd,
group_ndims=1,
n_samples=n_particles)
return variational
def q_net(x, z_dim, n_particles):
with zs.BayesianNet() as variational:
lz_x = tf.layers.dense(tf.to_float(x), 500, activation=tf.nn.relu)
lz_x = tf.layers.dense(lz_x, 500, activation=tf.nn.relu)
z_mean = tf.layers.dense(lz_x, z_dim)
z_logstd = tf.layers.dense(lz_x, z_dim)
z = zs.Normal('z',
z_mean,
logstd=z_logstd,
group_ndims=1,
n_samples=n_particles)
return variational
def q_net(observed, x, n_z, n_particles):
with zs.BayesianNet(observed=observed) as variational:
lz_x = tf.layers.dense(tf.to_float(x), 500, activation=tf.nn.relu)
lz_x = tf.layers.dense(lz_x, 500, activation=tf.nn.relu)
lz_mean = tf.layers.dense(lz_x, n_z)
lz_logstd = tf.layers.dense(lz_x, n_z)
z = zs.Normal('z',
lz_mean,
logstd=lz_logstd,
n_samples=n_particles,
group_ndims=1)
return variational
def build_q_net(x, z_dim, n_particles):
bn = zs.BayesianNet()
h = tf.layers.dense(tf.cast(x, tf.float32), 500, activation=tf.nn.relu)
h = tf.layers.dense(h, 500, activation=tf.nn.relu)
z_mean = tf.layers.dense(h, z_dim)
z_logstd = tf.layers.dense(h, z_dim)
bn.normal("z",
z_mean,
logstd=z_logstd,
group_ndims=1,
n_samples=n_particles)
return bn
def M2(observed, n, n_x, n_y, n_z, n_particles):
with zs.BayesianNet(observed=observed) as model:
z_mean = tf.zeros([n, n_z])
z = zs.Normal('z', z_mean, std=1., n_samples=n_particles,
group_ndims=1)
y_logits = tf.zeros([n, n_y])
y = zs.OnehotCategorical('y', y_logits, n_samples=n_particles)
lx_zy = layers.fully_connected(tf.concat([z, tf.to_float(y)], 2), 500)
lx_zy = layers.fully_connected(lx_zy, 500)
x_logits = layers.fully_connected(lx_zy, n_x, activation_fn=None)
x = zs.Bernoulli('x', x_logits, group_ndims=1)
return model
def gmm(observed, n, n_x, n_z):
with zs.BayesianNet(observed=observed) as model:
log_pi = tf.get_variable('log_pi', n_z, initializer=tf.truncated_normal_initializer(mean=1., stddev=0.5),
regularizer=var_regularizer(1.0))
mu = tf.get_variable('mu', [n_x, n_z], initializer=tf.orthogonal_initializer(gain=4.0)) # try uniform init
log_sigma = tf.get_variable('log_sigma', [n_x, n_z], initializer=tf.truncated_normal_initializer(stddev=0.5),
regularizer=l1_regularizer(0.01)) # try not l1_reg
z = zs.OnehotCategorical('z', log_pi, n_samples=n)
x_mean = tf.matmul(tf.to_float(z.tensor), tf.transpose(mu))
x_logstd = tf.matmul(tf.to_float(z.tensor), tf.transpose(log_sigma))
x = zs.Normal('x', x_mean, x_logstd, group_event_ndims=1)
return model, x.tensor, z.tensor
def build_gen(y, x_dim, z_dim, n, n_particles=1):
bn = zs.BayesianNet()
z_mean = tf.zeros([n, z_dim])
z = bn.normal("z", z_mean, std=1., group_ndims=1, n_samples=n_particles)
y = tf.cast(tf.repeat(tf.expand_dims(y, axis=0), repeats=n_particles, axis=0), tf.float32)
cat_z_y = tf.concat([z, y], axis=2)
h = tf.layers.dense(cat_z_y, 500, activation=tf.nn.relu)
h = tf.layers.dense(h, 500, activation=tf.nn.relu)
x_logits = tf.layers.dense(h, x_dim)
bn.deterministic("x_mean", tf.sigmoid(x_logits))
bn.bernoulli("x", x_logits, group_ndims=1)
return bn
def mean_field_variational(layer_sizes, n_particles):
with zs.BayesianNet() as variational:
ws = []
for i, (n_in, n_out) in enumerate(zip(layer_sizes[:-1], layer_sizes[1:])):
w_mean = tf.get_variable('w_mean_' + str(i), shape=[1, n_out, n_in + 1],initializer=tf.constant_initializer(0.))
w_logstd = tf.get_variable(
'w_logstd_' + str(i), shape=[1, n_out, n_in + 1],
initializer=tf.constant_initializer(0.))
ws.append(
zs.Normal('w' + str(i), w_mean, logstd=w_logstd,
n_samples=n_particles, group_ndims=2))
return variational
def q_net(observed, x, n_z, n_particles):
with zs.BayesianNet(observed=observed) as variational:
lz_x = layers.fully_connected(tf.to_float(x), 500)
lz_x = layers.fully_connected(lz_x, 500)
lz_mean = layers.fully_connected(lz_x, n_z, activation_fn=None)
lz_logstd = layers.fully_connected(lz_x, n_z, activation_fn=None)
z = zs.Normal('z',
lz_mean,
logstd=lz_logstd,
n_samples=n_particles,
group_event_ndims=1)
return variational
def qz_xy(x, y, n_z, n_particles):
with zs.BayesianNet() as variational:
lz_xy = layers.fully_connected(tf.to_float(tf.concat([x, y], 1)), 500)
lz_xy = layers.fully_connected(lz_xy, 500)
lz_mean = layers.fully_connected(lz_xy, n_z, activation_fn=None)
lz_logstd = layers.fully_connected(lz_xy, n_z, activation_fn=None)
z = zs.Normal('z',
lz_mean,
logstd=lz_logstd,
n_samples=n_particles,
group_event_ndims=1)
return variational
def unlabeled_proposal(x, n_y, n_z, n_particles):
with zs.BayesianNet() as proposal:
y_logits = qy_x(x, n_y)
y = zs.OnehotCategorical('y', y_logits, n_samples=n_particles)
x_tiled = tf.tile(tf.expand_dims(x, 0), [n_particles, 1, 1])
z_mean, z_logstd = qz_xy(x_tiled, y, n_z)
z = zs.Normal('z',
z_mean,
logstd=z_logstd,
group_event_ndims=1,
is_reparameterized=False)
return proposal
