本篇博客主要介绍TensorFlow中的name_scope和variable_scope。
name_scope示例代码：
import tensorflow as tf
with tf.name_scope('scope1'):
initializer = tf.constant_initializer(value=1)
var1 = tf.get_variable(name='var1', shape=[1], dtype=tf.float32, initializer=initializer)
var2 = tf.Variable(name='var2', initial_value=[2], dtype=tf.float32)
var21 = tf.Variable(name='var2', initial_value=[2.1], dtype=tf.float32)
var22 = tf.Variable(name='var2', initial_value=[2.2], dtype=tf.float32)
with tf.Session() as sess:
sess.run(tf.initialize_all_variables())
print(var1.name)
print(sess.run(var1))
print(var2.name)
print(sess.run(var2))
print(var21.name)
print(sess.run(var21))
print(var22.name)
print(sess.run(var22))
运行结果：
var1:0
[ 1.]
scope1/var2:0
[ 2.]
scope1/var2_1:0
[ 2.0999999]
scope1/var2_2:0
[ 2.20000005]结果解释：
tf.name_scope()对tf.get_variable()创建的变量没有作用，对于tf.Variable()创建的变量会产生作用。
对于相同名称的变量，TensorFlow会按照创建顺序在后面添加数字进行重新命名。
variable_scope示例代码：
import tensorflow as tf
with tf.variable_scope('scope1'):
initializer = tf.constant_initializer(value=1)
var1 = tf.get_variable(name='var1', shape=[1], dtype=tf.float32, initializer=initializer)
var2 = tf.Variable(name='var2', initial_value=[2], dtype=tf.float32)
var2_reuse = tf.Variable(name='var2', initial_value=[2], dtype=tf.float32)
with tf.Session() as sess:
sess.run(tf.initialize_all_variables())
print(var1.name)
print(sess.run(var1))
print(var2.name)
print(sess.run(var2))
print(var2_reuse.name)
print(sess.run(var2_reuse))
运行结果：
scope1/var1:0
[ 1.]
scope1/var2:0
[ 2.]
scope1/var2_1:0
[ 2.]结果解释：
tf.variable_scope()对tf.get_variable()创建的变量会产生作用，对tf.Variable()创建变量的重复使用会重新创建一个新的变量。
对于tf.get_variable创建变量的重复使用会报错。
with tf.variable_scope('scope1'):
initializer = tf.constant_initializer(value=1)
var1 = tf.get_variable(name='var1', shape=[1], dtype=tf.float32, initializer=initializer)
var1_reuse = tf.get_variable(name='var1')报错如下：
ValueError: Variable scope1/var1 already exists, disallowed. Did you mean to set reuse=True in VarScope? Originally defined at:重复使用的方法，声明scope.reuse_variables()
import tensorflow as tf
with tf.variable_scope('scope1') as scope:
initializer = tf.constant_initializer(value=1)
var1 = tf.get_variable(name='var1', shape=[1], dtype=tf.float32, initializer=initializer)
scope.reuse_variables()
var1_reuse = tf.get_variable(name='var1')
with tf.Session() as sess:
sess.run(tf.initialize_all_variables())
print(var1.name)
print(sess.run(var1))
print(var1_reuse.name)
print(sess.run(var1_reuse))
运行结果：
scope1/var1:0
[ 1.]
scope1/var1:0
[ 1.]
name_scope和variable_scope通常在RNN中使用到。
如rnn中test中重复使用train的参数：
# 22 scope (name_scope/variable_scope)
from __future__ import print_function
import tensorflow as tf
class TrainConfig:
batch_size = 20
time_steps = 20
input_size = 10
output_size = 2
cell_size = 11
learning_rate = 0.01
class TestConfig(TrainConfig):
time_steps = 1
class RNN(object):
def __init__(self, config):
self._batch_size = config.batch_size
self._time_steps = config.time_steps
self._input_size = config.input_size
self._output_size = config.output_size
self._cell_size = config.cell_size
self._lr = config.learning_rate
self._built_RNN()
def _built_RNN(self):
with tf.variable_scope('inputs'):
self._xs = tf.placeholder(tf.float32, [self._batch_size, self._time_steps, self._input_size], name='xs')
self._ys = tf.placeholder(tf.float32, [self._batch_size, self._time_steps, self._output_size], name='ys')
with tf.name_scope('RNN'):
with tf.variable_scope('input_layer'):
l_in_x = tf.reshape(self._xs, [-1, self._input_size], name='2_2D') # (batch*n_step, in_size)
# Ws (in_size, cell_size)
Wi = self._weight_variable([self._input_size, self._cell_size])
print(Wi.name)
# bs (cell_size, )
bi = self._bias_variable([self._cell_size, ])
# l_in_y = (batch * n_steps, cell_size)
with tf.name_scope('Wx_plus_b'):
l_in_y = tf.matmul(l_in_x, Wi) + bi
l_in_y = tf.reshape(l_in_y, [-1, self._time_steps, self._cell_size], name='2_3D')
with tf.variable_scope('cell'):
cell = tf.contrib.rnn.BasicLSTMCell(self._cell_size)
with tf.name_scope('initial_state'):
self._cell_initial_state = cell.zero_state(self._batch_size, dtype=tf.float32)
self.cell_outputs = []
cell_state = self._cell_initial_state
for t in range(self._time_steps):
if t > 0: tf.get_variable_scope().reuse_variables()
cell_output, cell_state = cell(l_in_y[:, t, :], cell_state)
self.cell_outputs.append(cell_output)
self._cell_final_state = cell_state
with tf.variable_scope('output_layer'):
# cell_outputs_reshaped (BATCH*TIME_STEP, CELL_SIZE)
cell_outputs_reshaped = tf.reshape(tf.concat(self.cell_outputs, 1), [-1, self._cell_size])
Wo = self._weight_variable((self._cell_size, self._output_size))
bo = self._bias_variable((self._output_size,))
product = tf.matmul(cell_outputs_reshaped, Wo) + bo
# _pred shape (batch*time_step, output_size)
self._pred = tf.nn.relu(product) # for displacement
with tf.name_scope('cost'):
_pred = tf.reshape(self._pred, [self._batch_size, self._time_steps, self._output_size])
mse = self.ms_error(_pred, self._ys)
mse_ave_across_batch = tf.reduce_mean(mse, 0)
mse_sum_across_time = tf.reduce_sum(mse_ave_across_batch, 0)
self._cost = mse_sum_across_time
self._cost_ave_time = self._cost / self._time_steps
with tf.variable_scope('trian'):
self._lr = tf.convert_to_tensor(self._lr)
self.train_op = tf.train.AdamOptimizer(self._lr).minimize(self._cost)
@staticmethod
def ms_error(y_target, y_pre):
return tf.square(tf.subtract(y_target, y_pre))
@staticmethod
def _weight_variable(shape, name='weights'):
initializer = tf.random_normal_initializer(mean=0., stddev=0.5, )
return tf.get_variable(shape=shape, initializer=initializer, name=name)
@staticmethod
def _bias_variable(shape, name='biases'):
initializer = tf.constant_initializer(0.1)
return tf.get_variable(name=name, shape=shape, initializer=initializer)
if __name__ == '__main__':
train_config = TrainConfig()
test_config = TestConfig()
# the wrong method to reuse parameters in train rnn
with tf.variable_scope('train_rnn'):
train_rnn1 = RNN(train_config)
with tf.variable_scope('test_rnn'):
test_rnn1 = RNN(test_config)
# the right method to reuse parameters in train rnn
with tf.variable_scope('rnn') as scope:
sess = tf.Session()
train_rnn2 = RNN(train_config)
scope.reuse_variables()
test_rnn2 = RNN(test_config)
# tf.initialize_all_variables() no long valid from
# 2017-03-02 if using tensorflow >= 0.12
if int((tf.__version__).split('.')[1]) < 12 and int((tf.__version__).split('.')[0]) < 1:
init = tf.initialize_all_variables()
else:
init = tf.global_variables_initializer()
sess.run(init)