def get_optimizer(model, opt, lr=None, adam_alpha=None, adam_beta1=None,
adam_beta2=None, adam_eps=None, weight_decay=None):
if opt == 'MomentumSGD':
optimizer = optimizers.MomentumSGD(lr=lr, momentum=0.9)
elif opt == 'Adam':
optimizer = optimizers.Adam(
alpha=adam_alpha, beta1=adam_beta1,
beta2=adam_beta2, eps=adam_eps)
elif opt == 'AdaGrad':
optimizer = optimizers.AdaGrad(lr=lr)
elif opt == 'RMSprop':
optimizer = optimizers.RMSprop(lr=lr)
else:
raise Exception('No optimizer is selected')
# The first model as the master model
optimizer.setup(model)
if opt == 'MomentumSGD':
optimizer.add_hook(
chainer.optimizer.WeightDecay(weight_decay))
return optimizer
python类RMSprop()的实例源码
def get_optimizer(opt, lr=None, adam_alpha=None, adam_beta1=None,
adam_beta2=None, adam_eps=None, weight_decay=None):
if opt == 'MomentumSGD':
optimizer = optimizers.MomentumSGD(lr=lr, momentum=0.9)
elif opt == 'Adam':
optimizer = optimizers.Adam(
alpha=adam_alpha, beta1=adam_beta1,
beta2=adam_beta2, eps=adam_eps)
elif opt == 'AdaGrad':
optimizer = optimizers.AdaGrad(lr=lr)
elif opt == 'RMSprop':
optimizer = optimizers.RMSprop(lr=lr)
else:
raise Exception('No optimizer is selected')
# The first model as the master model
if opt == 'MomentumSGD':
optimizer.decay = weight_decay
return optimizer
def get_optimizer(self, name, lr, momentum=0.9):
if name.lower() == "adam":
return optimizers.Adam(alpha=lr, beta1=momentum)
if name.lower() == "smorms3":
return optimizers.SMORMS3(lr=lr)
if name.lower() == "adagrad":
return optimizers.AdaGrad(lr=lr)
if name.lower() == "adadelta":
return optimizers.AdaDelta(rho=momentum)
if name.lower() == "nesterov" or name.lower() == "nesterovag":
return optimizers.NesterovAG(lr=lr, momentum=momentum)
if name.lower() == "rmsprop":
return optimizers.RMSprop(lr=lr, alpha=momentum)
if name.lower() == "momentumsgd":
return optimizers.MomentumSGD(lr=lr, mommentum=mommentum)
if name.lower() == "sgd":
return optimizers.SGD(lr=lr)
def get_optimizer(name, lr, momentum=0.9):
if name.lower() == "adam":
return optimizers.Adam(alpha=lr, beta1=momentum)
if name.lower() == "eve":
return Eve(alpha=lr, beta1=momentum)
if name.lower() == "adagrad":
return optimizers.AdaGrad(lr=lr)
if name.lower() == "adadelta":
return optimizers.AdaDelta(rho=momentum)
if name.lower() == "nesterov" or name.lower() == "nesterovag":
return optimizers.NesterovAG(lr=lr, momentum=momentum)
if name.lower() == "rmsprop":
return optimizers.RMSprop(lr=lr, alpha=momentum)
if name.lower() == "momentumsgd":
return optimizers.MomentumSGD(lr=lr, mommentum=mommentum)
if name.lower() == "sgd":
return optimizers.SGD(lr=lr)
def update_momentum(self, momentum):
if isinstance(self.optimizer, optimizers.Adam):
self.optimizer.beta1 = momentum
return
if isinstance(self.optimizer, Eve):
self.optimizer.beta1 = momentum
return
if isinstance(self.optimizer, optimizers.AdaDelta):
self.optimizer.rho = momentum
return
if isinstance(self.optimizer, optimizers.NesterovAG):
self.optimizer.momentum = momentum
return
if isinstance(self.optimizer, optimizers.RMSprop):
self.optimizer.alpha = momentum
return
if isinstance(self.optimizer, optimizers.MomentumSGD):
self.optimizer.mommentum = momentum
return
def get_optimizer(name, lr, momentum=0.9):
if name.lower() == "adam":
return chainer.optimizers.Adam(alpha=lr, beta1=momentum)
if name.lower() == "eve":
return Eve(alpha=lr, beta1=momentum)
if name.lower() == "adagrad":
return chainer.optimizers.AdaGrad(lr=lr)
if name.lower() == "adadelta":
return chainer.optimizers.AdaDelta(rho=momentum)
if name.lower() == "nesterov" or name.lower() == "nesterovag":
return chainer.optimizers.NesterovAG(lr=lr, momentum=momentum)
if name.lower() == "rmsprop":
return chainer.optimizers.RMSprop(lr=lr, alpha=momentum)
if name.lower() == "momentumsgd":
return chainer.optimizers.MomentumSGD(lr=lr, mommentum=mommentum)
if name.lower() == "sgd":
return chainer.optimizers.SGD(lr=lr)
raise Exception()
def update_momentum(self, momentum):
if isinstance(self.optimizer, optimizers.Adam):
self.optimizer.beta1 = momentum
return
if isinstance(self.optimizer, Eve):
self.optimizer.beta1 = momentum
return
if isinstance(self.optimizer, optimizers.AdaDelta):
self.optimizer.rho = momentum
return
if isinstance(self.optimizer, optimizers.NesterovAG):
self.optimizer.momentum = momentum
return
if isinstance(self.optimizer, optimizers.RMSprop):
self.optimizer.alpha = momentum
return
if isinstance(self.optimizer, optimizers.MomentumSGD):
self.optimizer.mommentum = momentum
return
def update_momentum(self, momentum):
if isinstance(self._optimizer, optimizers.Adam):
self._optimizer.beta1 = momentum
return
if isinstance(self._optimizer, Eve):
self._optimizer.beta1 = momentum
return
if isinstance(self._optimizer, optimizers.AdaDelta):
self._optimizer.rho = momentum
return
if isinstance(self._optimizer, optimizers.NesterovAG):
self._optimizer.momentum = momentum
return
if isinstance(self._optimizer, optimizers.RMSprop):
self._optimizer.alpha = momentum
return
if isinstance(self._optimizer, optimizers.MomentumSGD):
self._optimizer.mommentum = momentum
return
def get_optimizer(name, lr, momentum=0.9):
if name.lower() == "adam":
return optimizers.Adam(alpha=lr, beta1=momentum)
if name.lower() == "eve":
return Eve(alpha=lr, beta1=momentum)
if name.lower() == "adagrad":
return optimizers.AdaGrad(lr=lr)
if name.lower() == "adadelta":
return optimizers.AdaDelta(rho=momentum)
if name.lower() == "nesterov" or name.lower() == "nesterovag":
return optimizers.NesterovAG(lr=lr, momentum=momentum)
if name.lower() == "rmsprop":
return optimizers.RMSprop(lr=lr, alpha=momentum)
if name.lower() == "momentumsgd":
return optimizers.MomentumSGD(lr=lr, mommentum=mommentum)
if name.lower() == "sgd":
return optimizers.SGD(lr=lr)
def test_share_states(self):
model = L.Linear(2, 2)
opt_a = optimizers.RMSprop()
opt_a.setup(model)
arrays = async.share_states_as_shared_arrays(opt_a)
opt_b = optimizers.RMSprop()
opt_b.setup(copy.deepcopy(model))
# In Chainer v2, a model cannot be set up by two optimizers or more.
opt_c = optimizers.RMSprop()
opt_c.setup(copy.deepcopy(model))
"""
Removed the tests by assert_different_pointers
since they are trivial now.
"""
async.set_shared_states(opt_b, arrays)
async.set_shared_states(opt_c, arrays)
def assert_same_pointers(a, b):
a = a.target
b = b.target
for param_name, param_a in a.namedparams():
param_b = dict(b.namedparams())[param_name]
state_a = param_a.update_rule.state
state_b = param_b.update_rule.state
self.assertTrue(state_a)
self.assertTrue(state_b)
for state_name, state_val_a in state_a.items():
state_val_b = state_b[state_name]
self.assertTrue(isinstance(
state_val_a, np.ndarray))
self.assertTrue(isinstance(
state_val_b, np.ndarray))
self.assertEqual(state_val_a.ctypes.data,
state_val_b.ctypes.data)
assert_same_pointers(opt_a, opt_b)
assert_same_pointers(opt_a, opt_c)
def __init__(self, model=None, lr=0.045, decay=0.9, eps=1.0, weight_decay=4.0e-5, clip=2.0):
super(OptimizerGooglenetV3, self).__init__(model)
optimizer = optimizers.RMSprop(lr, decay, eps)
weight_decay = chainer.optimizer.WeightDecay(weight_decay)
clip = chainer.optimizer.GradientClipping(clip)
optimizer.setup(self.model)
optimizer.add_hook(weight_decay)
optimizer.add_hook(clip)
self.optimizer = optimizer
nutszebra_optimizer.py 文件源码
项目：neural_architecture_search_with_reinforcement_learning_appendix_a
作者: nutszebra
项目源码
文件源码
阅读 19
收藏 0
点赞 0
评论 0
def __init__(self, model=None, lr=0.045, decay=0.9, eps=1.0, weight_decay=4.0e-5, clip=2.0):
super(OptimizerGooglenetV3, self).__init__(model)
optimizer = optimizers.RMSprop(lr, decay, eps)
weight_decay = chainer.optimizer.WeightDecay(weight_decay)
clip = chainer.optimizer.GradientClipping(clip)
optimizer.setup(self.model)
optimizer.add_hook(weight_decay)
optimizer.add_hook(clip)
self.optimizer = optimizer
test_optimizers_by_linear_model.py 文件源码
项目：chainer-deconv
作者: germanRos
项目源码
文件源码
阅读 20
收藏 0
点赞 0
评论 0
def create(self):
return optimizers.RMSprop(0.1)
def __init__(self, model=None, lr=0.045, decay=0.9, eps=1.0, weight_decay=4.0e-5, clip=2.0):
super(OptimizerGooglenetV3, self).__init__(model)
optimizer = optimizers.RMSprop(lr, decay, eps)
weight_decay = chainer.optimizer.WeightDecay(weight_decay)
clip = chainer.optimizer.GradientClipping(clip)
optimizer.setup(self.model)
optimizer.add_hook(weight_decay)
optimizer.add_hook(clip)
self.optimizer = optimizer
nutszebra_optimizer.py 文件源码
项目：pyramidal_residual_networks
作者: nutszebra
项目源码
文件源码
阅读 19
收藏 0
点赞 0
评论 0
def __init__(self, model=None, lr=0.045, decay=0.9, eps=1.0, weight_decay=4.0e-5, clip=2.0):
super(OptimizerGooglenetV3, self).__init__(model)
optimizer = optimizers.RMSprop(lr, decay, eps)
weight_decay = chainer.optimizer.WeightDecay(weight_decay)
clip = chainer.optimizer.GradientClipping(clip)
optimizer.setup(self.model)
optimizer.add_hook(weight_decay)
optimizer.add_hook(clip)
self.optimizer = optimizer
def train(args):
nz = args.nz
batch_size = args.batch_size
epochs = args.epochs
gpu = args.gpu
# CIFAR-10 images in range [-1, 1] (tanh generator outputs)
train, _ = datasets.get_cifar10(withlabel=False, ndim=3, scale=2)
train -= 1.0
train_iter = iterators.SerialIterator(train, batch_size)
z_iter = RandomNoiseIterator(GaussianNoiseGenerator(0, 1, args.nz),
batch_size)
optimizer_generator = optimizers.RMSprop(lr=0.00005)
optimizer_critic = optimizers.RMSprop(lr=0.00005)
optimizer_generator.setup(Generator())
optimizer_critic.setup(Critic())
updater = WassersteinGANUpdater(
iterator=train_iter,
noise_iterator=z_iter,
optimizer_generator=optimizer_generator,
optimizer_critic=optimizer_critic,
device=gpu)
trainer = training.Trainer(updater, stop_trigger=(epochs, 'epoch'))
trainer.extend(extensions.ProgressBar())
trainer.extend(extensions.LogReport(trigger=(1, 'iteration')))
trainer.extend(GeneratorSample(), trigger=(1, 'epoch'))
trainer.extend(extensions.PrintReport(['epoch', 'iteration', 'critic/loss',
'critic/loss/real', 'critic/loss/fake', 'generator/loss']))
trainer.run()
def __init__(self, model=None, lr=0.045, decay=0.9, eps=1.0, weight_decay=4.0e-5, clip=2.0):
super(OptimizerGooglenetV3, self).__init__(model)
optimizer = optimizers.RMSprop(lr, decay, eps)
weight_decay = chainer.optimizer.WeightDecay(weight_decay)
clip = chainer.optimizer.GradientClipping(clip)
optimizer.setup(self.model)
optimizer.add_hook(weight_decay)
optimizer.add_hook(clip)
self.optimizer = optimizer
def __init__(self, model=None, lr=0.045, decay=0.9, eps=1.0, weight_decay=4.0e-5, clip=2.0):
super(OptimizerGooglenetV3, self).__init__(model)
optimizer = optimizers.RMSprop(lr, decay, eps)
weight_decay = chainer.optimizer.WeightDecay(weight_decay)
clip = chainer.optimizer.GradientClipping(clip)
optimizer.setup(self.model)
optimizer.add_hook(weight_decay)
optimizer.add_hook(clip)
self.optimizer = optimizer
