def optimizer(self):
lr = tf.get_variable('learning_rate', initializer=5e-4, trainable=False)
opt = tf.train.AdamOptimizer(lr, epsilon=1e-3)
return optimizer.apply_grad_processors(
opt, [
gradproc.ScaleGradient(('STN.*', 0.1)),
gradproc.SummaryGradient()])
def optimizer(self):
lr = tf.get_variable('learning_rate',
initializer=self.learning_rate,
trainable=False)
opt = tf.train.RMSPropOptimizer(lr, epsilon=1e-5)
return optimizer.apply_grad_processors(opt,
[gradproc.SummaryGradient()])
def optimizer(self):
opt = tf.train.AdamOptimizer(self.cfg.learning_rate)
return optimizer.apply_grad_processors(opt, [
gradproc.MapGradient(
lambda grad: tf.clip_by_average_norm(grad, 0.3)),
gradproc.SummaryGradient()
])
def optimizer(self):
lr = tf.get_variable("learning_rate", initializer=0.0002, trainable=False)
tf.summary.scalar("learning_rate", lr)
opt = tf.train.AdamOptimizer(lr, beta1=0.5, beta2=0.999)
return optimizer.apply_grad_processors(
opt, [gradproc.SummaryGradient(), gradproc.CheckGradient()]
)
def _get_optimizer(self):
lr = tf.get_variable('learning_rate',
initializer=1e-3,
trainable=False)
opt = tf.train.AdamOptimizer(lr, epsilon=1e-3)
return optimizer.apply_grad_processors(
opt, [gradproc.GlobalNormClip(10),
gradproc.SummaryGradient()])
def optimizer(self):
lr = tf.get_variable('learning_rate', initializer=self.learning_rate, trainable=False)
# opt = tf.train.AdamOptimizer(lr, epsilon=1e-3)
opt = tf.train.AdamOptimizer(lr)
return optimizer.apply_grad_processors(
opt, [
# gradproc.GlobalNormClip(2.0),
gradproc.MapGradient(lambda grad: tf.clip_by_average_norm(grad, 0.5)),
gradproc.SummaryGradient()])
def optimizer(self):
lr = tf.train.exponential_decay(self.base_lr,
global_step=get_global_step_var(),
decay_steps=self.decay_steps,
decay_rate=self.decay_rate,
name='learning-rate')
opt = tf.train.RMSPropOptimizer(learning_rate=lr)
tf.summary.scalar('lr', lr)
return optimizer.apply_grad_processors(opt,
[gradproc.SummaryGradient()])
def optimizer(self):
lr = tf.get_variable('learning_rate',
initializer=1e-3,
trainable=False)
tf.summary.scalar("learning_rate", lr)
opt = tf.train.RMSPropOptimizer(lr,
decay=0.95,
momentum=0.95,
epsilon=1e-2)
return optimizer.apply_grad_processors(opt,
[gradproc.SummaryGradient()])
def optimizer(self):
lr = tf.get_variable('learning_rate', initializer=1e-3, trainable=False)
# This will also put the summary in tensorboard, stat.json and print in terminal,
# but this time without moving average
tf.summary.scalar('lr', lr)
# opt = tf.train.MomentumOptimizer(lr, 0.9)
opt = tf.train.AdamOptimizer(lr)
return optimizer.apply_grad_processors(
opt, [gradproc.MapGradient(lambda grad: tf.clip_by_average_norm(grad, 0.5)),
gradproc.SummaryGradient()])
def _get_optimizer(self):
lr = symbf.get_scalar_var('learning_rate', 1e-3, summary=True)
opt = tf.train.AdamOptimizer(lr, epsilon=1e-3)
return optimizer.apply_grad_processors(
opt, [gradproc.GlobalNormClip(10),
gradproc.SummaryGradient()])
def optimizer(self):
lr = tf.get_variable('learning_rate', initializer=1e-3, trainable=False)
# opt = tf.train.MomentumOptimizer(lr, 0.9)
opt = tf.train.AdamOptimizer(learning_rate=lr)
# return opt
return optimizer.apply_grad_processors(opt, [gradproc.ScaleGradient(('5_Tr_Cv.*', 0.1)), gradproc.SummaryGradient()])