本文整理汇总了Python中layer.Layer.build方法的典型用法代码示例。如果您正苦于以下问题：Python Layer.build方法的具体用法？Python Layer.build怎么用？Python Layer.build使用的例子？那么恭喜您, 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layer.Layer的用法示例。
在下文中一共展示了Layer.build方法的1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: NeuralNetwork
# 需要导入模块: from layer import Layer [as 别名]
# 或者: from layer.Layer import build [as 别名]
class NeuralNetwork(object):
"""
A Neural network class with a scikit-compliant interface
"""
cost_functions = ['log-likelihood', 'cross-entropy']
regularizations = ['l1', 'l2', 'none']
## --------------------------------------------
def __init__(
self,
hidden_layers = [],
learning_algorithm='Adam',
output_activation='softmax',
cost_function='cross-entropy',
regularization='none',
reg_lambda=1.0,
learning_rate=0.1,
early_stopping=False,
stagnation=10,
n_epochs=10,
mini_batch_size=10,
):
"""
Constructor
"""
## List of Layer objects, the input and output layers are automatically specified
## using the input data and the output_activation parameter
self.hidden_layers = hidden_layers
## Parameters from the constructor
self.learning_algorithm = learning_algorithm
assert cost_function in self.cost_functions, 'Available cost functions are {0}'.format(', '.join(self.cost_functions))
self.cost_function = cost_function
self.output_activation = output_activation
assert regularization in self.regularizations, 'Available regularizations are {0}'.format(', '.join(self.regularizations))
self.regularization = regularization
self.reg_lambda = reg_lambda
self.learning_rate = learning_rate
self.early_stopping = early_stopping
self.stagnation = stagnation
self.n_epochs = n_epochs
self.mini_batch_size = mini_batch_size
## --------------------------------------------
def build(self, X, y):
"""
Builds the neural network in tensorflow
"""
## Start a tensorflow interactive session
self.session = tf.InteractiveSession()
## First, create a placeholder for the targets
self.targets = tf.placeholder(tf.float32, shape=[None, self.n_categories])
## First, create the input layer
self.input_layer = Layer(n_neurons=self.n_features)
self.input_layer.build()
## Then create all the hidden layers
current_input_layer = self.input_layer
for layer in self.hidden_layers:
layer.build(current_input_layer)
current_input_layer = layer
## Create the output layer
self.output_layer = Layer(n_neurons=self.n_categories, activation=self.output_activation)
self.output_layer.build(current_input_layer)
## Define the cost function
self.cost = None
if self.cost_function == 'log-likelihood':
self.cost = tf.reduce_mean(-tf.log(tf.reduce_sum(self.targets * self.output_layer.output, reduction_indices=[1])))
else:
self.cost = tf.reduce_mean(-tf.reduce_sum(self.targets * tf.log(self.output_layer.output), reduction_indices=[1]))
## Define the regularization parameters and function
self.reg_lambda_param = tf.placeholder(tf.float32)
self.batch_size = tf.placeholder(tf.float32)
if self.regularization == 'l1':
self.reg_term = tf.reduce_sum(tf.abs(self.output_layer.weights))
for layer in self.hidden_layers:
self.reg_term += tf.reduce_sum(tf.abs(layer.weights))
elif self.regularization == 'l2':
self.reg_term = tf.reduce_sum(self.output_layer.weights * self.output_layer.weights)
for layer in self.hidden_layers:
self.reg_term += tf.reduce_sum(layer.weights * layer.weights)
#.........这里部分代码省略.........