TensorFlow Computational Graph Explained

In TensorFlow, the computation graph is a kind of data flow graph that describes the relationships between data flow and operations. It is made up of nodes and edges, where nodes represent operations and edges represent data flow.

To use a computational graph, you need to first create a default graph. You can create a new graph using tf.Graph(), or get the default graph using tf.get_default_graph(). Then, you can use a with statement to add operations to the graph. For example:

import tensorflow as tf

# 创建一个新的计算图
graph = tf.Graph()

# 将操作添加到计算图中
with graph.as_default():
    a = tf.constant(3)
    b = tf.constant(4)
    c = a + b

# 创建一个会话
with tf.Session(graph=graph) as sess:
    result = sess.run(c)
    print(result)  # 输出：7

In the example above, we first create a new computational graph, then add constants a and b as well as the addition operation c to the graph, and use a session to execute the operations in the graph. TensorFlow automatically builds the computational graph and performs the operations, resulting in an output of 7.