def _build_graph(self, images, labels, mode):
"""Constructs the TF graph for the cifar model.
Args:
images: A 4-D image Tensor
labels: A 2-D labels Tensor.
mode: string indicating training mode ( e.g., 'train', 'valid', 'test').
"""
is_training = 'train' in mode
if is_training:
self.global_step = tf.train.get_or_create_global_step()
logits = build_model(images, self.num_classes, is_training,
self.hparams)
self.predictions, self.cost = helper_utils.setup_loss(logits, labels)
self.accuracy, self.eval_op = tf.metrics.accuracy(
tf.argmax(labels, 1), tf.argmax(self.predictions, 1))
self._calc_num_trainable_params()
# Adds L2 weight decay to the cost
self.cost = helper_utils.decay_weights(self.cost,
self.hparams.weight_decay_rate)
if is_training:
self._build_train_op()
# Setup checkpointing for this child model
# Keep 2 or more checkpoints around during training.
with tf.device('/cpu:0'):
self.saver = tf.train.Saver(max_to_keep=2)
self.init = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
def _build_graph(self, images, labels, mode):
"""Constructs the TF graph for the cifar model.
Args:
images: A 4-D image Tensor
labels: A 2-D labels Tensor.
mode: string indicating training mode ( e.g., 'train', 'valid', 'test').
"""
is_training = 'train' in mode
if is_training:
self.global_step = tf.train.get_or_create_global_step()
if self.hparams.use_gamma_swish:
for layer in range(13):
_ = tf.Variable([self.hparams.init_beta],
trainable=True,
dtype=tf.float32,
name='swish_beta_layer_{}'.format(layer))
_ = tf.Variable([self.hparams.init_gamma],
trainable=True,
dtype=tf.float32,
name='swish_gamma_layer_{}'.format(layer))
logits, hiddens = build_model(images, self.num_classes, is_training,
self.hparams)
self.predictions, self.cost_ = helper_utils.setup_loss(logits, labels)
self.hiddens_norm = tf.reduce_mean(hiddens**2)
self.logits = logits
self.logit_norm = tf.reduce_mean(logits**2)
self.accuracy, self.eval_op = tf.metrics.accuracy(
tf.argmax(labels, 1), tf.argmax(self.predictions, 1))
self._calc_num_trainable_params()
self.compute_flops_per_example()
# Adds L2 weight decay to the cost
self.cost = helper_utils.decay_weights(self.cost_,
self.hparams.weight_decay_rate)
if is_training:
self._build_train_op()
# Setup checkpointing for this child model
# Keep 2 or more checkpoints around during training.
with tf.device('/cpu:0'):
self.saver = tf.train.Saver(max_to_keep=2)
self.ckpt_saver = tf.train.Saver(max_to_keep=100)
self.init = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
def _build_graph(self, images, labels, mode):
"""Constructs the TF graph for the cifar model.
Args:
images: A 4-D image Tensor
labels: A 2-D labels Tensor.
mode: string indicating training mode ( e.g., 'train', 'valid', 'test').
"""
is_training = 'train' in mode
if is_training:
self.global_step = tf.train.get_or_create_global_step()
logits = build_model(
images,
self.num_classes,
is_training,
self.hparams)
self.predictions, self.cost = helper_utils.setup_loss(
logits, labels)
self.accuracy, self.eval_op = tf.metrics.accuracy(
tf.argmax(labels, 1), tf.argmax(self.predictions, 1))
self._calc_num_trainable_params()
# Adds L2 weight decay to the cost
self.cost = helper_utils.decay_weights(self.cost,
self.hparams.weight_decay_rate)
if is_training:
self._build_train_op()
# Setup checkpointing for this child model
# Keep 2 or more checkpoints around during training.
with tf.device('/cpu:0'):
self.saver = tf.train.Saver(max_to_keep=2)
self.init = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
