def test_basic_ops_value(self):
np.random.seed(12082518)
x = K.variable(np.random.randn(8, 8))
y = K.variable(np.random.randn(8, 8))
z = K.variable(np.random.randint(0, 2, size=(8, 8)), dtype=np.bool)
w = K.variable(np.random.randint(0, 2, size=(8, 8)), dtype=np.bool)
self.assertEqual(round(np.sum(K.eval(K.relu(x, alpha=0.12))) * 10000),
276733)
self.assertEqual(round(np.sum(K.eval(K.elu(x, alpha=0.12))) * 10000),
289202)
self.assertEqual(np.sum(K.eval(K.softmax(x))), 8.0)
self.assertEqual(round(np.sum(K.eval(K.softplus(x))) * 10000), 554564)
self.assertEqual(round(np.sum(K.eval(K.softsign(x))) * 100000), 211582)
self.assertEqual(round(np.sum(K.eval(K.sigmoid(x))) * 10000), 330427)
self.assertEqual(round(np.sum(K.eval(K.hard_sigmoid(x))) * 10000),
330836)
self.assertEqual(round(np.sum(K.eval(K.tanh(x))) * 100000), 290165)
self.assertEqual(round(np.sum(K.eval(K.square(x))) * 10000), 744492)
self.assertEqual(round(np.sum(K.eval(K.sqrt(x))) * 10000), 300212)
self.assertEqual(round(np.sum(K.eval(K.abs(x))) * 10000), 559979)
self.assertEqual(np.sum(K.eval(K.sign(x))), 6.0)
self.assertEqual(round(np.sum(K.eval(K.inv(x))) * 1000), 495838)
self.assertEqual(round(np.sum(K.eval(K.exp(x))) * 1000), 122062)
self.assertEqual(round(np.sum(K.eval(K.log(K.abs(x)))) * 10000),
-344491)
self.assertEqual(np.sum(K.eval(K.round(x))), 5.0)
self.assertEqual(round(np.sum(K.eval(K.pow(x, 8))) * 100), 398153)
self.assertEqual(
round(np.sum(K.eval(K.clip(x, -0.12, 0.12))) * 1000000), 620529)
# TODO: pygpu (libgpuarray) still not support diag
# self.assertEqual(round(np.sum(K.eval(K.diag(x))) * 100000), 325289)
self.assertEqual(np.sum(K.eval(K.eye(12, 8))), 8.0)
self.assertEqual(np.sum(K.eval(K.eq(z, w))), 38)
self.assertEqual(np.sum(K.eval(K.neq(z, w))), 26)
self.assertEqual(np.sum(K.eval(K.gt(x, y))), 33)
self.assertEqual(np.sum(K.eval(K.ge(x, y))), 33)
self.assertEqual(np.sum(K.eval(K.lt(x, y))), 31)
self.assertEqual(np.sum(K.eval(K.le(x, y))), 31)
self.assertEqual(round(np.sum(K.eval(K.switch(z, x, y))) * 100000),
139884)
def convolutional_vae(X, saved_states, **kwargs):
""" convolutional_vae
Return
------
[y_encoder, y_decoder]
States
------
[f_inference (encoder), f_generative (decoder)]
"""
n = kwargs.get('n', 10)
batch_size = K.get_shape(X)[0]
if batch_size is None:
raise ValueError("You must specify batch_size dimension for the input placeholder.")
# ====== init ====== #
if saved_states is None:
# Encoder
f_inference = N.Sequence([
N.Reshape(shape=(-1, 28, 28, 1)),
N.Conv(num_filters=32, filter_size=3, strides=1, pad='valid',
b_init=init_ops.constant_initializer(0.), activation=K.elu),
N.Conv(num_filters=64, filter_size=5, strides=2, pad='same',
b_init=init_ops.constant_initializer(0.), activation=K.elu),
N.Dropout(level=0.1),
N.Flatten(outdim=2),
N.Dense(num_units=n * 2, b_init=None),
N.BatchNorm(axes=0)
], debug=True, name='Encoder')
# Decoder
f_generative = N.Sequence([
N.Dimshuffle(pattern=(0, 'x', 'x', 1)),
N.TransposeConv(num_filters=64, filter_size=3, strides=1, pad='valid',
b_init=init_ops.constant_initializer(0.), activation=K.elu),
N.TransposeConv(num_filters=32, filter_size=5, strides=2, pad='same',
b_init=init_ops.constant_initializer(0.), activation=K.elu),
N.TransposeConv(num_filters=1, filter_size=13, strides=3, pad='valid',
b_init=None),
N.BatchNorm(activation=K.linear),
N.Flatten(outdim=3)
], debug=True, name="Decoder")
else:
f_inference, f_generative = saved_states
# ====== Perfrom ====== #
# Encoder
y_encoder = f_inference(K.cast(X, 'float32'))
mu = y_encoder[:, :n]
sigma = K.softplus(y_encoder[:, n:])
qz = Normal(mu=mu, sigma=sigma, name='Normal_qz')
# Decoder
z = Normal(mu=K.zeros(shape=(batch_size, n)),
sigma=K.ones(shape=(batch_size, n)), name="Normal_pz")
logits = f_generative(z)
X_reconstruct = Bernoulli(logits=logits)
# inference
params = f_inference.parameters + f_generative.parameters
inference = ed.KLqp(latent_vars={z: qz}, data={X_reconstruct: X})
# ====== get cost for training ====== #
# Bind p(x, z) and q(z | x) to the same placeholder for x.
if K.is_training():
import tensorflow as tf
inference.initialize()
if True:
optimizer = tf.train.AdamOptimizer(0.01, epsilon=1.0)
updates = optimizer.apply_gradients(
optimizer.compute_gradients(inference.loss, var_list=params))
init = tf.global_variables_initializer()
init.run()
f_train = K.function(X, inference.loss, updates)
else:
optimizer = tf.train.AdamOptimizer(0.01, epsilon=1.0)
inference.initialize(optimizer=optimizer, var_list=params)
init = tf.global_variables_initializer()
init.run()
f_train = lambda x: inference.update(feed_dict={X: x})['loss']
samples = K.sigmoid(logits)
return (samples, z, qz), (f_inference, f_generative)
def convolutional_vae(X, saved_states, **kwargs):
""" convolutional_vae
Return
------
[y_encoder, y_decoder]
States
------
[f_inference (encoder), f_generative (decoder)]
"""
n = kwargs.get('n', 10)
batch_size = K.get_shape(X)[0]
if batch_size is None:
raise ValueError(
"You must specify batch_size dimension for the input placeholder.")
# ====== init ====== #
if saved_states is None:
# Encoder
f_inference = N.Sequence([
N.Reshape(shape=(-1, 28, 28, 1)),
N.Conv(num_filters=32,
filter_size=3,
strides=1,
pad='valid',
b_init=init_ops.constant_initializer(0.),
activation=K.elu),
N.Conv(num_filters=64,
filter_size=5,
strides=2,
pad='same',
b_init=init_ops.constant_initializer(0.),
activation=K.elu),
N.Dropout(level=0.1),
N.Flatten(outdim=2),
N.Dense(num_units=n * 2, b_init=None),
N.BatchNorm(axes=0)
],
debug=True,
name='Encoder')
# Decoder
f_generative = N.Sequence([
N.Dimshuffle(pattern=(0, 'x', 'x', 1)),
N.TransposeConv(num_filters=64,
filter_size=3,
strides=1,
pad='valid',
b_init=init_ops.constant_initializer(0.),
activation=K.elu),
N.TransposeConv(num_filters=32,
filter_size=5,
strides=2,
pad='same',
b_init=init_ops.constant_initializer(0.),
activation=K.elu),
N.TransposeConv(num_filters=1,
filter_size=13,
strides=3,
pad='valid',
b_init=None),
N.BatchNorm(activation=K.linear),
N.Flatten(outdim=3)
],
debug=True,
name="Decoder")
else:
f_inference, f_generative = saved_states
# ====== Perfrom ====== #
# Encoder
y_encoder = f_inference(K.cast(X, 'float32'))
mu = y_encoder[:, :n]
sigma = K.softplus(y_encoder[:, n:])
qz = Normal(mu=mu, sigma=sigma, name='Normal_qz')
# Decoder
z = Normal(mu=K.zeros(shape=(batch_size, n)),
sigma=K.ones(shape=(batch_size, n)),
name="Normal_pz")
logits = f_generative(z)
X_reconstruct = Bernoulli(logits=logits)
# inference
params = f_inference.parameters + f_generative.parameters
inference = ed.KLqp(latent_vars={z: qz}, data={X_reconstruct: X})
# ====== get cost for training ====== #
# Bind p(x, z) and q(z | x) to the same placeholder for x.
if K.is_training():
import tensorflow as tf
inference.initialize()
if True:
optimizer = tf.train.AdamOptimizer(0.01, epsilon=1.0)
updates = optimizer.apply_gradients(
optimizer.compute_gradients(inference.loss, var_list=params))
init = tf.global_variables_initializer()
init.run()
f_train = K.function(X, inference.loss, updates)
else:
optimizer = tf.train.AdamOptimizer(0.01, epsilon=1.0)
inference.initialize(optimizer=optimizer, var_list=params)
init = tf.global_variables_initializer()
init.run()
f_train = lambda x: inference.update(feed_dict={X: x})['loss']
samples = K.sigmoid(logits)
return (samples, z, qz), (f_inference, f_generative)