def __init__(self,
relu=True,
init_zero=False,
center=True,
scale=True,
data_format='channels_last',
**kwargs):
super(BatchNormRelu, self).__init__(**kwargs)
self.relu = relu
if init_zero:
gamma_initializer = tf.zeros_initializer()
else:
gamma_initializer = tf.ones_initializer()
if data_format == 'channels_first':
axis = 1
else:
axis = -1
self.bn = tf.keras.layers.BatchNormalization(
axis=axis,
momentum=BATCH_NORM_DECAY,
epsilon=BATCH_NORM_EPSILON,
center=center,
scale=scale,
fused=False,
gamma_initializer=gamma_initializer)
def test_shared_sequence_non_sequence_into_input_layer(self):
non_seq = tf.feature_column.categorical_column_with_identity(
'non_seq', num_buckets=10)
seq = tf.feature_column.sequence_categorical_column_with_identity(
'seq', num_buckets=10)
shared_non_seq, shared_seq = tf.feature_column.shared_embeddings(
[non_seq, seq],
dimension=4,
combiner='sum',
initializer=tf.ones_initializer(),
shared_embedding_collection_name='shared')
seq = tf.SparseTensor(indices=[[0, 0], [0, 1], [1, 0]],
values=[0, 1, 2],
dense_shape=[2, 2])
non_seq = tf.SparseTensor(indices=[[0, 0], [0, 1], [1, 0]],
values=[0, 1, 2],
dense_shape=[2, 2])
features = {'seq': seq, 'non_seq': non_seq}
# Tile the context features across the sequence features
seq_input, seq_length = ksfc.SequenceFeatures([shared_seq])(features)
non_seq_input = dense_features.DenseFeatures([shared_non_seq
])(features)
with self.cached_session() as sess:
sess.run(tf.compat.v1.global_variables_initializer())
output_seq, output_seq_length, output_non_seq = sess.run(
[seq_input, seq_length, non_seq_input])
self.assertAllEqual(
output_seq,
[[[1, 1, 1, 1], [1, 1, 1, 1]], [[1, 1, 1, 1], [0, 0, 0, 0]]])
self.assertAllEqual(output_seq_length, [2, 1])
self.assertAllEqual(output_non_seq, [[2, 2, 2, 2], [1, 1, 1, 1]])
def __init__(self,
num_groups=None,
group_size=None,
eps=1e-5,
beta_init=tf.zeros_initializer(),
gamma_init=tf.ones_initializer(),
**kwargs):
"""Initializer.
Args:
num_groups: int, the number of channel-groups to normalize over.
group_size: int, size of the groups to normalize over.
eps: float, a small additive constant to avoid /sqrt(0).
beta_init: initializer for bias, defaults to zeros.
gamma_init: initializer for scale, defaults to ones.
**kwargs: other tf.keras.layers.Layer arguments.
"""
super(GroupNormalization, self).__init__(**kwargs)
if num_groups is None and group_size is None:
num_groups = 32
self._num_groups = num_groups
self._group_size = group_size
self._eps = eps
self._beta_init = beta_init
self._gamma_init = gamma_init
def __init__(self, hdim, dtype=tf.float32, name="LayerNorm"):
super(NormLayer, self).__init__(name=name)
self._dtype = dtype
with tf.compat.v1.variable_scope(name):
self.beta = tf.compat.v1.get_variable(
"beta", [hdim], dtype=dtype, initializer=tf.zeros_initializer())
self.gamma = tf.compat.v1.get_variable(
"gamma", [hdim], dtype=dtype, initializer=tf.ones_initializer())
def __init__(self,
relu=True,
init_zero=False,
center=True,
scale=True,
data_format='channels_last',
**kwargs):
super(BatchNormRelu, self).__init__(**kwargs)
self.relu = relu
if init_zero:
gamma_initializer = tf.zeros_initializer()
else:
gamma_initializer = tf.ones_initializer()
if data_format == 'channels_first':
axis = 1
else:
axis = -1
if FLAGS.global_bn:
# TODO(srbs): Set fused=True
# Batch normalization layers with fused=True only support 4D input
# tensors.
self.bn = tf.keras.layers.experimental.SyncBatchNormalization(
axis=axis,
momentum=FLAGS.batch_norm_decay,
epsilon=BATCH_NORM_EPSILON,
center=center,
scale=scale,
gamma_initializer=gamma_initializer)
else:
# TODO(srbs): Set fused=True
# Batch normalization layers with fused=True only support 4D input
# tensors.
self.bn = tf.keras.layers.BatchNormalization(
axis=axis,
momentum=FLAGS.batch_norm_decay,
epsilon=BATCH_NORM_EPSILON,
center=center,
scale=scale,
fused=False,
gamma_initializer=gamma_initializer)
def call(self, input_tensor):
inputs = tf.convert_to_tensor(input_tensor)
inputs_shape = get_shape_list(inputs)
inputs_rank = len(inputs_shape)
dtype = inputs.dtype.base_dtype
norm_axis = inputs_rank - 1
params_shape = [inputs_shape[norm_axis]]
# Allocate parameters for the beta and gamma of the normalization.
if self.beta is None:
self.beta = tf.compat.v1.get_variable(
"beta",
shape=params_shape,
dtype=dtype,
initializer=tf.zeros_initializer(),
trainable=True)
self._trainable_weights.append(self.beta)
if self.gamma is None:
self.gamma = tf.compat.v1.get_variable(
"gamma",
shape=params_shape,
dtype=dtype,
initializer=tf.ones_initializer(),
trainable=True)
self._trainable_weights.append(self.gamma)
# Compute norm along last axis
mean, variance = tf.nn.moments(inputs, [norm_axis], keepdims=True)
# Compute layer normalization using the batch_normalization function.
# Note that epsilon must be increased for float16 due to the limited
# representable range.
variance_epsilon = 1e-12 if dtype != tf.float16 else 1e-3
outputs = tf.nn.batch_normalization(inputs,
mean,
variance,
offset=self.beta,
scale=self.gamma,
variance_epsilon=variance_epsilon)
outputs.set_shape(inputs_shape)
return outputs
def test_generate_candidates(self,
want_names,
want_subnetwork_losses,
want_mixture_weight_losses,
want_complexities,
learn_mixture_weights=False,
initial_num_layers=0,
previous_ensemble=None):
feature_columns = [tf.feature_column.numeric_column("x")]
generator = simple_dnn.Generator(
feature_columns=feature_columns,
optimizer=tf.compat.v1.train.GradientDescentOptimizer(.1),
layer_size=3,
initial_num_layers=initial_num_layers,
learn_mixture_weights=learn_mixture_weights,
seed=42)
with context.graph_mode(), tf.Graph().as_default() as g:
iteration_step = tf.compat.v1.train.create_global_step()
features = {"x": [[1.], [2.]]}
labels = tf.constant([[0.], [1.]])
names = []
subnetwork_losses = []
mixture_weight_losses = []
complexities = []
for builder in generator.generate_candidates(
previous_ensemble,
# The following arguments are not used by
# simple_dnn.BuilderGenerator's generate_candidates.
iteration_number=0,
previous_ensemble_reports=[],
all_reports=[]):
names.append(builder.name)
# 1. Build subnetwork graph.
subnetwork = builder.build_subnetwork(
features,
logits_dimension=1,
training=True,
iteration_step=iteration_step,
summary=tf.summary,
previous_ensemble=previous_ensemble)
# 2. Build subnetwork train ops.
subnetwork_loss = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(
logits=subnetwork.logits, labels=labels))
subnetwork_train_op = builder.build_subnetwork_train_op(
subnetwork,
subnetwork_loss,
var_list=None,
labels=labels,
iteration_step=iteration_step,
summary=tf.summary,
previous_ensemble=None)
# 3. Build mixture weight train ops.
# Stop gradients since mixture weights should have not propagate
# beyond top layer.
subnetwork_logits = tf.stop_gradient(subnetwork.logits)
# Mixture weight will initialize to a one-valued scalar.
mixture_weight_logits = tf.compat.v1.layers.dense(
subnetwork_logits,
units=1,
use_bias=False,
kernel_initializer=tf.ones_initializer())
mixture_weight_loss = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(
logits=mixture_weight_logits, labels=labels))
mixture_weight_train_op = builder.build_mixture_weights_train_op(
mixture_weight_loss,
var_list=None,
labels=labels,
logits=mixture_weight_logits,
iteration_step=iteration_step,
summary=tf.summary)
with self.test_session(graph=g) as sess:
sess.run(tf.compat.v1.global_variables_initializer())
sess.run(subnetwork_train_op)
sess.run(mixture_weight_train_op)
subnetwork_losses.append(sess.run(subnetwork_loss))
mixture_weight_losses.append(sess.run(mixture_weight_loss))
complexities.append(sess.run(subnetwork.complexity))
self.assertEqual(want_names, names)
self.assertAllClose(want_subnetwork_losses,
subnetwork_losses,
atol=1e-3)
self.assertAllClose(want_mixture_weight_losses,
mixture_weight_losses,
atol=1e-3)
self.assertAllClose(want_complexities, complexities, atol=1e-3)