tensorflow 中 name_scope和variable_scope

from http://blog.csdn.net/appleml/article/details/53668237

1. import tensorflow as tf
2. with tf.name_scope("hello") as name_scope:
3. arr1 = tf.get_variable("arr1", shape=[2,10],dtype=tf.float32)
4. print (name_scope)
5. print (arr1.name)
6. print ("scope_name:%s " % tf.get_variable_scope().original_name_scope)

运行后的结果如下：

hello/

arr1:0

scope_name:

1. import tensorflow as tf
2. with tf.name_scope('hidden') as scope:
3. a = tf.constant(5, name='alpha')
4. W = tf.Variable(tf.random_uniform([1, 2], -1.0, 1.0), name='weights')
5. b = tf.Variable(tf.zeros([1]), name='biases')
6. print (a.name)
7. print (W.name)
8. print (b.name)

运行的结果：

hidden/alpha:0

hidden/weights:0

hidden/biases:0

红色字体要强调的部分所以把字体改成了红色，理解name_scope 对 tf.get_variable()的作用和 tf.Variable()的不同