我们从Python开源项目中，提取了以下50个代码示例，用于说明如何使用chainer.optimizers.MomentumSGD()。
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
def __init__(self, model=None, lr=0.1, momentum=0.9, schedule=(250, 375), weight_decay=1.0e-4): super(OptimizerStochasticDepth, self).__init__(model) self.lr = lr self.momentum = momentum self.schedule = schedule self.weight_decay = weight_decay all_links = OptimizerStochasticDepth._find(model) optimizer_set = [] for link in all_links: optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(self.weight_decay) optimizer.setup(link[0]) optimizer.add_hook(weight_decay) optimizer_set.append(optimizer) self.optimizer_set = optimizer_set self.all_links = all_links
def __init__(self, model=None, lr=0.1, momentum=0.9, schedule=(250, 375), weight_decay=1.0e-4): super(OptimizerResnetOfResnet, self).__init__(model) self.lr = lr self.momentum = momentum self.schedule = schedule self.weight_decay = weight_decay all_links = OptimizerStochasticDepth._find(model) optimizer_set = [] for link in all_links: optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(self.weight_decay) optimizer.setup(link[0]) optimizer.add_hook(weight_decay) optimizer_set.append(optimizer) self.optimizer_set = optimizer_set self.all_links = all_links
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 test_remove_link(self): opt = optimizers.MomentumSGD(lr=0.01) # Update each depth for depth in six.moves.range(1, self.n_encdec + 1): model = segnet.SegNet(self.n_encdec, self.n_classes, self.x_shape[1], self.n_mid) model = segnet.SegNetLoss( model, class_weight=None, train_depth=depth) opt.setup(model) # Deregister non-target links from opt if depth > 1: model.predictor.remove_link('conv_cls') for d in range(1, self.n_encdec + 1): if d != depth: model.predictor.remove_link('encdec{}'.format(d)) for name, link in model.namedparams(): if depth > 1: self.assertTrue( 'encdec{}'.format(depth) in name) else: self.assertTrue( 'encdec{}'.format(depth) in name or 'conv_cls' in name)
def decay_learning_rate(opt, factor, final_value): if isinstance(opt, optimizers.NesterovAG): if opt.lr <= final_value: return final_value opt.lr *= factor return if isinstance(opt, optimizers.SGD): if opt.lr <= final_value: return final_value opt.lr *= factor return if isinstance(opt, optimizers.MomentumSGD): if opt.lr <= final_value: return final_value opt.lr *= factor return if isinstance(opt, optimizers.Adam): if opt.alpha <= final_value: return final_value opt.alpha *= factor return raise NotImplementedError()
def __init__(self, model=None, lr=0.02, momentum=0.9, schedule=(200, 300, 350, 375), weight_decay=1.0e-4): super(OptimizerFractalNet, self).__init__(model) self.lr = lr self.momentum = momentum self.schedule = schedule all_links = OptimizerFractalNet._find(model) optimizer_set = [] for link in all_links: wd = chainer.optimizer.WeightDecay(weight_decay) optimizer = optimizers.MomentumSGD(lr, momentum) optimizer.setup(link[0]) optimizer.add_hook(wd) optimizer_set.append(optimizer) self.optimizer_set = optimizer_set self.all_links = all_links self.flag = False
def __init__(self, model=None, lr=0.1, momentum=0.9, schedule=(250, 375), weight_decay=1.0e-4): super(OptimizerStochasticDepth, self).__init__(model) self.lr = lr self.momentum = momentum self.schedule = schedule self.weight_decay = weight_decay all_links = OptimizerStochasticDepth._find(model) optimizer_set = [] for link in all_links: optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(self.weight_decay) optimizer.setup(link[0]) optimizer.add_hook(weight_decay) optimizer_set.append(optimizer) self.optimizer_set = optimizer_set self.all_links = all_links self.flag = False
def __init__(self, model=None, lr=0.1, momentum=0.9, schedule=(250, 375), weight_decay=1.0e-4): super(OptimizerResnetOfResnet, self).__init__(model) self.lr = lr self.momentum = momentum self.schedule = schedule self.weight_decay = weight_decay all_links = OptimizerStochasticDepth._find(model) optimizer_set = [] for link in all_links: optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(self.weight_decay) optimizer.setup(link[0]) optimizer.add_hook(weight_decay) optimizer_set.append(optimizer) self.optimizer_set = optimizer_set self.all_links = all_links self.flag = False
def __init__(self, model=None, lr=0.5, momentum=0.9, schedule=(150, 225), weight_decay=1.0e-4): super(OptimizerPyramidalResNetWithSSD, self).__init__(model) self.lr = lr self.momentum = momentum self.schedule = schedule self.weight_decay = weight_decay all_links = OptimizerPyramidalResNetWithSSD._find(model) optimizer_set = [] for link in all_links: optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(self.weight_decay) optimizer.setup(link[0]) optimizer.add_hook(weight_decay) optimizer_set.append(optimizer) self.optimizer_set = optimizer_set self.all_links = all_links self.flag = False
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 decrease_learning_rate(opt, factor, final_value): if isinstance(opt, optimizers.NesterovAG): if opt.lr <= final_value: return final_value opt.lr *= factor return if isinstance(opt, optimizers.SGD): if opt.lr <= final_value: return final_value opt.lr *= factor return if isinstance(opt, optimizers.MomentumSGD): if opt.lr <= final_value: return final_value opt.lr *= factor return if isinstance(opt, optimizers.Adam): if opt.alpha <= final_value: return final_value opt.alpha *= factor return raise NotImplementedError()
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 __init__(self, model=None, schedule=(int(32000. / (50000. / 128)), int(48000. / (50000. / 128))), lr=0.1, momentum=0.9, weight_decay=1.0e-4, warm_up_lr=0.01): super(OptimizerResnet, self).__init__(model) optimizer = optimizers.MomentumSGD(warm_up_lr, momentum) weight_decay = chainer.optimizer.WeightDecay(weight_decay) optimizer.setup(self.model) optimizer.add_hook(weight_decay) self.optimizer = optimizer self.schedule = schedule self.lr = lr self.warmup_lr = warm_up_lr self.momentum = momentum self.weight_decay = weight_decay
def __init__(self, model=None, schedule=(150, 225), lr=0.1, momentum=0.9, weight_decay=1.0e-4): super(OptimizerDense, self).__init__(model) optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(weight_decay) optimizer.setup(self.model) optimizer.add_hook(weight_decay) self.optimizer = optimizer self.schedule = schedule self.lr = lr self.momentum = momentum self.weight_decay = weight_decay
def __init__(self, model=None, schedule=(60, 120, 160), lr=0.1, momentum=0.9, weight_decay=5.0e-4): super(OptimizerWideRes, self).__init__(model) optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(weight_decay) optimizer.setup(self.model) optimizer.add_hook(weight_decay) self.optimizer = optimizer self.schedule = schedule self.lr = lr self.momentum = momentum self.weight_decay = weight_decay
def __init__(self, model=None, schedule=(196, 224), lr=0.1, momentum=0.9, weight_decay=1.0e-4): super(OptimizerSwapout, self).__init__(model) optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(weight_decay) optimizer.setup(self.model) optimizer.add_hook(weight_decay) self.optimizer = optimizer self.schedule = schedule self.lr = lr self.momentum = momentum self.weight_decay = weight_decay
def __init__(self, model=None, lr=0.045, momentum=0.9, weight_decay=1.0e-5, period=2): super(OptimizerXception, self).__init__(model) optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(weight_decay) optimizer.setup(self.model) optimizer.add_hook(weight_decay) self.optimizer = optimizer self.lr = lr self.momentum = momentum self.weight_decay = weight_decay self.period = int(period)
def __init__(self, model=None, lr=0.0015, momentum=0.9, weight_decay=2.0e-4): super(OptimizerGooglenet, self).__init__(model) optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(weight_decay) optimizer.setup(self.model) optimizer.add_hook(weight_decay) self.optimizer = optimizer self.lr = lr self.momentum = momentum self.weight_decay = weight_decay
def __init__(self, model=None, lr=0.1, momentum=0.9, weight_decay=1.0e-4, schedule=(int(1.0e5 / (50000. / 128)), )): super(OptimizerNetworkInNetwork, self).__init__(model) optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(weight_decay) optimizer.setup(self.model) optimizer.add_hook(weight_decay) self.optimizer = optimizer self.lr = lr self.momentum = momentum self.weight_decay = weight_decay self.schedule = schedule
def __init__(self, model=None, lr=0.045, momentum=0.9, weight_decay=4.0e-5): super(OptimizerGooglenetV2, self).__init__(model) optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(weight_decay) optimizer.setup(self.model) optimizer.add_hook(weight_decay) self.optimizer = optimizer self.lr = lr self.momentum = momentum self.weight_decay = weight_decay
def __init__(self, model=None, lr=0.045, weight_decay=4.0e-5, clip=2.0): super(OptimizerGooglenetV3, self).__init__(model) optimizer = optimizers.MomentumSGD(lr, 0.9) 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=0.1, momentum=0.9, weight_decay=5.0e-4, schedule=(150, 225)): super(OptimizerResNext, self).__init__(model) optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(weight_decay) optimizer.setup(self.model) optimizer.add_hook(weight_decay) self.optimizer = optimizer self.schedule = schedule self.lr = lr self.momentum = momentum self.weight_decay = weight_decay
def __init__(self, model=None, lr=0.02, momentum=0.9, schedule=(150, 225, 300, 375)): super(OptimizerFractalNet, self).__init__(model) optimizer = optimizers.MomentumSGD(lr, momentum) optimizer.setup(self.model) self.optimizer = optimizer self.schedule = schedule self.lr = lr self.momentum = momentum
def __init__(self, model=None, lr=0.01, momentum=0.9, schedule=(150, 225), weight_decay=5.0e-4): super(OptimizerPReLUNet, self).__init__(model) optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(weight_decay) optimizer.setup(self.model) optimizer.add_hook(weight_decay) self.optimizer = optimizer self.schedule = schedule self.lr = lr self.momentum = momentum
def __init__(self, model=None, schedule=(42, 62), lr=0.1, momentum=0.9, weight_decay=1.0e-4): super(OptimizerResnetInResnet, self).__init__(model) optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(weight_decay) optimizer.setup(self.model) optimizer.add_hook(weight_decay) self.optimizer = optimizer self.schedule = schedule self.lr = lr self.momentum = momentum self.weight_decay = weight_decay
def __init__(self, model=None, schedule=(150, 175), lr=0.1, momentum=0.9, weight_decay=1.0e-4): super(OptimizerAppendixA, self).__init__(model) optimizer = optimizers.MomentumSGD(lr, momentum) weight_decay = chainer.optimizer.WeightDecay(weight_decay) optimizer.setup(self.model) optimizer.add_hook(weight_decay) self.optimizer = optimizer self.schedule = schedule self.lr = lr self.momentum = momentum self.weight_decay = weight_decay
