Tensorflow global_step TypeError

Question

我将Tensorboard添加到现有的小Tensorflow项目中，我知道该项目可以练习使用Tensorboard，但是我得到一个类型错误，global_step必须是字符串或张量，但是我已经将global_step分配给tf.Variable(0, name='global_step', trainable=False)就像文档和我在网上看到的每个例子一样。对我所缺少的任何想法都会非常感激。

---＆GT; 70 [summary_merge，tf.train.global_step（sess，global_step_tensor），update_model，weights]，feed_dict）

TypeError：Fetch参数0具有无效类型，必须是字符串或Tensor。 （无法将int转换为Tensor或Operation。）

import gym
import numpy as np
import random
import ipdb
import matplotlib.pyplot as plt
import tensorflow as tf
%matplotlib inline

tf.reset_default_graph()
env = gym.make('FrozenLake-v0')

global_step_tensor = tf.Variable(0, name='global_step', trainable=False)
saver = tf.train.Saver()

with tf.name_scope('policy-network'):
    observations = tf.placeholder(shape=[1, env.observation_space.n], dtype=tf.float32, name='input')
    weights = tf.Variable(tf.random_uniform([16,4], 0, 0.01))
    Qout = tf.matmul(observations, weights)
    predict = tf.argmax(Qout, 1)

    nextQ = tf.placeholder(shape=[1, 4], dtype=tf.float32)
    loss = tf.reduce_sum(tf.square(nextQ - Qout))
    trainer = tf.train.GradientDescentOptimizer(learning_rate=0.1)
    update_model = trainer.minimize(loss, global_step= global_step_tensor)

with tf.name_scope('summaries'):
    summary_merge = tf.summary.merge([
        tf.summary.histogram('loss', loss),
        tf.summary.histogram('weights', weights)
    ])


init = tf.global_variables_initializer()

gamma = .99
epsilon = 0.1
num_episodes = 2000

j_list = []
r_list = []

with tf.Session() as sess:
    # Create a writer for summary data
    writer = tf.summary.FileWriter('/tmp/display', sess.graph)
    sess.run(init)
    for i in range(num_episodes):
        s = env.reset()
        reward_all = 0
        done = False
        j = 0

        while j < 99:
            j+= 1

            a, allQ = sess.run([predict, Qout], feed_dict={observations:np.identity(16)[s:s+1]})
            if np.random.rand(1) < epsilon:
                a[0] = env.action_space.sample()

            s1, reward, done, _ = env.step(a[0])
            Q1 = sess.run(Qout, feed_dict={observations: np.identity(16)[s1:s1+1]}) 

            maxQ1 = np.max(Q1)
            targetQ = allQ

            targetQ[0, a[0]] = reward + gamma*maxQ1
            feed_dict = {observations: np.identity(16)[s:s+1], nextQ:targetQ}

            # Run prediction operation, evaluate summary_merge and assign to summaries
            summaries, global_step_tensor, _, W1 = sess.run(
            [summary_merge, tf.train.global_step(sess, global_step_tensor) ,update_model, weights], feed_dict)

            # Write summaries to writer
            writer.add_summary(summaries, global_step)

            reward_all += reward
            s = s1

            if done == True:
                epsilon = 1./((i/50) + 10)
                break

        if (i+1) % 1000 == 0:
            # Save the graph every 1000 episodes
            saver.save(sess, '/tmp/checkpoint', global_step=global_step)

        j_list.append(j)
        r_list.append(reward_all)

print("Percent of succesful episodes: {} %".format((sum(r_list)/num_episodes)))