tensorflow 牛刀小试，模型框架搭建

import tensorflow as tf

graph = tf.Graph()

arr = [i for i in range(20)]
with graph.as_default():
    sess = tf.Session(graph=graph)
    
    with tf.name_scope("variables"):
        global_steps = tf.Variable(0, dtype=tf.int32, trainable=False, name="global_steps")

        
    with tf.name_scope("input"):
        a = tf.placeholder(tf.int32, shape=[], name="placeholder_a")  #转化为张量

        
    with tf.name_scope("update"):
        increment_step = global_steps.assign_add(1)

    
    #注意这里，summary.scaler接受的参数是标量，如果是一个张量数组或者带维度的张量，可以使用tf.reduce_sum()降维，变为标量
    with tf.name_scope("summary"):
        output = a
        #output = tf.reduce_sum(a)
        tf.summary.scalar("output", output)


    with tf.name_scope("global_ops"):
        init = tf.global_variables_initializer()
        merged_summaries = tf.summary.merge_all()

        
    writer = tf.summary.FileWriter("./train", sess.graph)
    sess.run(init)


    def run_graph(input_tensor):
        feed_dict = {a: input_tensor} #字典的值可以使数字、字符串、列表、numpy，当shape=[]，则为数字
        step, summary = sess.run([increment_step, merged_summaries], feed_dict = feed_dict)
        writer.add_summary(summary, global_step=step)


    for i in range(len(arr)):
        input_tensor = i
        run_graph(input_tensor)

在代码中存在五个命名空间，但是显示时只有四个，我记得一种解释是tensorboard并不会将tf.summary显示出来，而在本模块并没有进行任何张量操作，所以不显示。

修改代码：

在summaries命名空间添加节点计算

import tensorflow as tf

graph = tf.Graph()

arr = [i for i in range(20)]
with graph.as_default():
    sess = tf.Session(graph=graph)
    with tf.name_scope("variables"):
        global_steps = tf.Variable(0, dtype=tf.int32, trainable=False, name="global_steps")

    with tf.name_scope("input"):
        a = tf.placeholder(tf.int32, shape=[], name="placeholder_a")

    with tf.name_scope("update"):
        increment_step = global_steps.assign_add(1)

    with tf.name_scope("summaries"):
        b = tf.constant(dtype=tf.int32, shape=[], value=2)
        output = a + b
        #tf.summary.scalar("output", output)
        tf.summary.scalar("output", a)

    with tf.name_scope("global_ops"):
        init = tf.global_variables_initializer()
        merged_summaries = tf.summary.merge_all()

    writer = tf.summary.FileWriter("./train", sess.graph)
    sess.run(init)


    def run_graph(input_tensor):
        feed_dict = {a: input_tensor}
        step, summary = sess.run([increment_step, merged_summaries], feed_dict = feed_dict)
        writer.add_summary(summary, global_step=step)


    for i in range(len(arr)):
        input_tensor = i
        run_graph(input_tensor)

则summaries可以显示出来。 

对于像这种简单的不追求模型复用（呗做成模块来给其他人用），可以按照这种框架来进行搭建