#### 1. 在会话(session)中计算图
```python
g = tf.Graph()
with tf.Session(graph=g) as sess:
    sess.run(...)
```
#### 2. 加载模型后不用给变量初始化
```python
with tf.Session(graph=g) as sess:
    # # initialize the variables
    # init = tf.global_variables_initializer()
    # sess.run(init)
    # create a saver to save variables in restoring
    saver = tf.train.Saver(tf.global_variables())
    # load the model
    saver.restore(sess, tf.train.latest_checkpoint(modelPath))
```
#### 3. 可同时使用多个session
```python
sess1 = tf.Session(g1)
sess2 = tf.Session(g2)
sess1.run(...)
sess2.run(...)
```
#### 但是，图计算过程中，(代码)左侧有相同变量在两个图中同时计算
```python
g1=tf.Graph()
g2=tf.Graph()
with g1.as_default():
    a = tf.constant([1.0, 1.0])
    b = tf.constant([1.0, 1.0])
    result1 = a+c
with g2.as_default():
    a = tf.constant([2.0, 2.0])
    b = tf.constant([2.0, 2.0])
    result2 = a+c
sess1 = tf.Session(graph=g1)
sess2 = tf.Session(graph=g2)
sess1.run(result1); sess1.run(result2) #正确
sess1.run(a) #错误
sess2.run(a) #正确
# 按顺序执行下来，a,b都应当在g2中计算
```
#### tf.gloabl_variables()得到的是某个graph的变量，需在graph的上下文管理器中运行
```python
with g.as_default():
    tf.global_variables()
```
#### tf.global_variables()---tensorflow, g.get_operations()---graph