TFcaiCH3

1. 定义两个不同的图

import tensorflow as tfg1 = tf.Graph()with g1.as_default():    v = tf.get_variable("v", [1], initializer = tf.zeros_initializer()) # 设置初始值为0g2 = tf.Graph()with g2.as_default():    v = tf.get_variable("v", [1], initializer = tf.ones_initializer())  # 设置初始值为1with tf.Session(graph = g1) as sess:    tf.global_variables_initializer().run()    with tf.variable_scope("", reuse=True):        print(sess.run(tf.get_variable("v")))with tf.Session(graph = g2) as sess:    tf.global_variables_initializer().run()    with tf.variable_scope("", reuse=True):        print(sess.run(tf.get_variable("v")))

[ 0.][ 1.]

2. 张量的概念

import tensorflow as tfa = tf.constant([1.0, 2.0], name="a")b = tf.constant([2.0, 3.0], name="b")result = a + bprint resultsess = tf.InteractiveSession ()print(result.eval())sess.close()

Tensor("add:0", shape=(2,), dtype=float32)[ 3.  5.]

3. 会话的使用

3.1 创建和关闭会话

# 创建一个会话。sess = tf.Session()# 使用会话得到之前计算的结果。print(sess.run(result))# 关闭会话使得本次运行中使用到的资源可以被释放。sess.close()

[ 3.  5.]

3.2 使用with statement 来创建会话

with tf.Session() as sess:    print(sess.run(result))

[ 3.  5.]

3.3 指定默认会话

sess = tf.Session()with sess.as_default():     print(result.eval())

[ 3.  5.]

sess = tf.Session()# 下面的两个命令有相同的功能。print(sess.run(result))print(result.eval(session=sess))

[ 3.  5.][ 3.  5.]

4. 使用tf.InteractiveSession构建会话

sess = tf.InteractiveSession ()print(result.eval())sess.close()

[ 3.  5.]

5. 通过ConfigProto配置会话

config=tf.ConfigProto(allow_soft_placement=True, log_device_placement=True)sess1 = tf.InteractiveSession(config=config)sess2 = tf.Session(config=config)