InvalidArgumentError：您必须使用dtype bool输入占位符张量“占位符”的值... [[accuracy / accuracy_var / _3191]]

Question

我已经用MNIST数据集训练了一个挤压和激励卷积神经网络模型。我已经将训练好的模型保存在一个文件夹中。

为了在生产中使用经过训练的模型，我已将其还原并尝试对新数据进行预测，但出现以下错误消息。

  

tensorflow.python.framework.errors_impl.InvalidArgumentError：您必须   用dtype bool输入占位符张量'Placeholder'的值          [[{{node占位符}}]          [[accuracy / accuracy_var / _3191]]

模型还原代码：

import tensorflow as tf
from cifar10 import *

init_learning_rate = 0.1
train_x, train_y, test_x, test_y = prepare_data()
train_x, test_x = color_preprocessing(train_x, test_x)
tf.reset_default_graph()
saver = tf.train.import_meta_graph("./model/ResNeXt.ckpt.meta")

with tf.Session() as sess:
    ckpt = tf.train.get_checkpoint_state('./model')
    if ckpt and tf.train.checkpoint_exists(ckpt.model_checkpoint_path):
        saver.restore(sess, ckpt.model_checkpoint_path)
        graph = tf.get_default_graph()

        for n in graph.as_graph_def().node:
            print(n.name)

        cost = graph.get_tensor_by_name("cost_reduce_mean:0")
        accuracy = graph.get_tensor_by_name("accuracy/accuracy_var:0")

        test_batch_x = test_x[0: 1000]
        test_batch_y = test_y[0: 1000]

        x = graph.get_tensor_by_name("placeholder_x_input:0")
        label = graph.get_tensor_by_name("placeholder_label_input:0")

        training_flag = tf.placeholder(tf.bool, name='tr_fg_99999')
        learning_rate = tf.placeholder(tf.float32, name='learning_rate_99999')

        test_feed_dict = {x: test_batch_x,label: test_batch_y,learning_rate: 0.1,training_flag: False}
        loss_, acc_ = sess.run([cost, accuracy], feed_dict=test_feed_dict)
        total_acc = acc_

行-loss_, acc_ = sess.run([cost, accuracy], feed_dict=test_feed_dict)一旦执行，就会出现错误。