tensorflow中出现{TypeError}unhashable type: 'numpy.ndarray'

本人实验中使用feed的方式填充数据，sess处的代码如下：

with tf.Session() as sess:
    init = tf.global_variables_initializer()
    sess.run(init)
    for epoch in range(a.epochs):
        input, target = load_batch_data(batch_size=16, a=a)
        batch_input = input.astype(np.float32)
        batch_target = target.astype(np.float32)
        sess.run(predict_real, feed_dict={input: batch_input, target: batch_target})

运行的时候出现：{TypeError}unhashable type: 'numpy.ndarray'

后　　来　　发　　现：

在session外边定义input和target的时候是这么写的：

input = tf.placeholder(dtype=tf.float32, shape=[None, image_size, image_size, num_channels])
target = tf.placeholder(dtype=tf.float32, shape=[None, image_size, image_size, num_channels])

然而，我在开启session后又定义了input，target。这导致我在运行下面这行代码的时候，

sess.run(predict_real, feed_dict={input: batch_input, target: batch_target})

出现了{TypeError}unhashable type: 'numpy.ndarray'这样的错误。然而此input和target非session外面的input和target。知道是这个原因后，改正的话就很简单了，修改session内input和target的名称即可，如下：

    with tf.Session() as sess:
        init = tf.global_variables_initializer()
        sess.run(init)
        if a.mode == 'train':
            for epoch in range(a.epochs):
                batch_input, batch_target = load_batch_data(a=a)
                batch_input = batch_input.astype(np.float32)
                batch_target = batch_target.astype(np.float32)
                sess.run(model, feed_dict={input: batch_input, target: batch_target})
                print('epoch' + str(epoch) + ':')
            saver.save(sess, 'model_parameter/train.ckpt')
            print('training finished!!!')
        elif a.mode == 'test':
            #ceshi
            ckpt = tf.train.latest_checkpoint(a.checkpoint)
            saver.restore(sess, ckpt)
            # 获取测试时候的图像，然后添加标签
            batch_input, _ = load_batch_data(a=a)
            # batch_input = batch_input / 255.
            batch_input = batch_input.astype(np.float32)
            generator_output = sess.run(test_output, feed_dict={input: batch_input})
            # 对结果进行处理，图像通道上减去3，得到rgb图像
            result = process_generator_output(generator_output)
            if result:
                print('测试完成！')
        else:
            print('the MODE is not avaliable...')