本文整理汇总了Python中lasagne.layers.get_all_params方法的典型用法代码示例。如果您正苦于以下问题:Python layers.get_all_params方法的具体用法?Python layers.get_all_params怎么用?Python layers.get_all_params使用的例子?那么, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类lasagne.layers
的用法示例。
在下文中一共展示了layers.get_all_params方法的5个代码示例,这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞,您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: net_updates
# 需要导入模块: from lasagne import layers [as 别名]
# 或者: from lasagne.layers import get_all_params [as 别名]
def net_updates(net, loss, lr):
# Get all trainable parameters (weights) of our net
params = l.get_all_params(net, trainable=True)
# We use the adam update, other options are available
if cfg.OPTIMIZER == 'adam':
param_updates = updates.adam(loss, params, learning_rate=lr, beta1=0.9)
elif cfg.OPTIMIZER == 'nesterov':
param_updates = updates.nesterov_momentum(loss, params, learning_rate=lr, momentum=0.9)
elif cfg.OPTIMIZER == 'sgd':
param_updates = updates.sgd(loss, params, learning_rate=lr)
return param_updates
#################### TRAIN FUNCTION #####################
示例2: make_training_functions
# 需要导入模块: from lasagne import layers [as 别名]
# 或者: from lasagne.layers import get_all_params [as 别名]
def make_training_functions(network, new_params, input_var, aug_var, target_var):
output = lasagne.layers.get_output(network, deterministic=True, batch_norm_use_averages=True, batch_norm_update_averages=False);
loss = lasagne.objectives.binary_crossentropy(output, target_var).mean();
deter_output = lasagne.layers.get_output(network, deterministic=True);
deter_loss = lasagne.objectives.binary_crossentropy(deter_output, target_var).mean();
params = layers.get_all_params(network, trainable=True);
updates = lasagne.updates.nesterov_momentum(loss, params, learning_rate=LearningRate, momentum=0.985);
new_params_updates = lasagne.updates.nesterov_momentum(loss, new_params, learning_rate=LearningRate, momentum=0.985);
val_fn = theano.function([input_var, aug_var, target_var], [deter_loss, deter_output]);
train_fn = theano.function([input_var, aug_var, target_var], loss, updates=updates);
new_params_train_fn = theano.function([input_var, aug_var, target_var], loss, updates=new_params_updates);
return train_fn, new_params_train_fn, val_fn;
示例3: make_training_functions
# 需要导入模块: from lasagne import layers [as 别名]
# 或者: from lasagne.layers import get_all_params [as 别名]
def make_training_functions(network, new_params, input_var, aug_var, target_var):
output = lasagne.layers.get_output(network);
loss = lasagne.objectives.binary_crossentropy(output, target_var).mean();
deter_output = lasagne.layers.get_output(network, deterministic=True);
deter_loss = lasagne.objectives.binary_crossentropy(deter_output, target_var).mean();
params = layers.get_all_params(network, trainable=True);
updates = lasagne.updates.nesterov_momentum(loss, params, learning_rate=LearningRate, momentum=0.985);
new_params_updates = lasagne.updates.nesterov_momentum(loss, new_params, learning_rate=LearningRate, momentum=0.985);
val_fn = theano.function([input_var, aug_var, target_var], [deter_loss, deter_output]);
train_fn = theano.function([input_var, aug_var, target_var], loss, updates=updates);
new_params_train_fn = theano.function([input_var, aug_var, target_var], loss, updates=new_params_updates);
return train_fn, new_params_train_fn, val_fn;
示例4: make_training_functions
# 需要导入模块: from lasagne import layers [as 别名]
# 或者: from lasagne.layers import get_all_params [as 别名]
def make_training_functions(network, new_params, input_var, aug_var, target_var):
output = lasagne.layers.get_output(network, deterministic=False);
loss = lasagne.objectives.binary_crossentropy(output, target_var).mean();
deter_output = lasagne.layers.get_output(network, deterministic=True);
deter_loss = lasagne.objectives.binary_crossentropy(deter_output, target_var).mean();
params = layers.get_all_params(network, trainable=True);
updates = lasagne.updates.nesterov_momentum(loss, params, learning_rate=LearningRate, momentum=0.985);
new_params_updates = lasagne.updates.nesterov_momentum(loss, new_params, learning_rate=LearningRate, momentum=0.985);
val_fn = theano.function([input_var, aug_var, target_var], [deter_loss, deter_output]);
train_fn = theano.function([input_var, aug_var, target_var], loss, updates=updates);
new_params_train_fn = theano.function([input_var, aug_var, target_var], loss, updates=new_params_updates);
return train_fn, new_params_train_fn, val_fn;
示例5: make_training_functions
# 需要导入模块: from lasagne import layers [as 别名]
# 或者: from lasagne.layers import get_all_params [as 别名]
def make_training_functions(network, new_params, input_var, aug_var, target_var):
output = lasagne.layers.get_output(network, deterministic=True);
loss = lasagne.objectives.binary_crossentropy(output, target_var).mean();
deter_output = lasagne.layers.get_output(network, deterministic=True);
deter_loss = lasagne.objectives.binary_crossentropy(deter_output, target_var).mean();
params = layers.get_all_params(network, trainable=True);
updates = lasagne.updates.nesterov_momentum(loss, params, learning_rate=LearningRate, momentum=0.985);
new_params_updates = lasagne.updates.nesterov_momentum(loss, new_params, learning_rate=LearningRate, momentum=0.985);
val_fn = theano.function([input_var, aug_var, target_var], [deter_loss, deter_output]);
train_fn = theano.function([input_var, aug_var, target_var], loss, updates=updates);
new_params_train_fn = theano.function([input_var, aug_var, target_var], loss, updates=new_params_updates);
return train_fn, new_params_train_fn, val_fn;