TensorFlow 学习（三）—— Variables（tf.initialize_all_variables()/tf.global_variables_initializer()）

0. tf.initialize_all_variables()/tf.global_variables_initializer()

What are the differences between tf.initialize_all_variables() and tf.global_variables_initializer()

注意对于 tf.initialize_all_variables() 接口，TensorFlow 文档有一个重要说明：

tf.initialize_all_variables(): THIS FUNCTION IS DEPRECATED. It will be removed after 2017-03-02. Instructions for updating: Use tf.global_variables_initializer instead.

tf.initialize_all_variables() 该函数将不再使用，在 2017年3月2号以后；

用 tf.global_variables_initializer() 替代 tf.initialize_all_variables()

1. 变量初始化

变量初始化的标准形式：

init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)

当然也可简写为：

tf.Session().run(tf.initialize_all_variables())

如何有选择地初始化部分变量呢？使用

tf.initialize_variables()

，比如要初始化

v_6, v_7, v_8

三个变量：

init_new_vars_op = tf.initialize_variables([v_6, v_7, v_8])
sess.run(init_new_vars_op)

2. 识别未被初始化的变量

用 try & except 语句块捕获：

uninit_vars = []
for var in tf.all_variables():
try:
sess.run(var)
except tf.errors.FailedPreconditionError:
uninit_vars.append(var)

init_new_vars_op = tf.initialize_variables(uninit_vars)

In TensorFlow is there any way to just initialize uninitialised variables?

3. 变量的更新

>> state = tf.Variable(1, name='counter')
>> add_one = tf.add(state, tf.constant(1))
>> update = tf.assign(state, add_one)
>> with tf.Session() as sess:
sess.run(tf.gloabl_variables_initializer())
sess.run(state)
for _ in range(3):
sess.run(update)
print(sess.run(state))