Tensorflow - eval() 错误：您必须为占位符张量提供一个值

我尝试使用 eval() 来了解每个学习步骤中发生的情况。

但是，如果我在 tf.matmul 操作上使用 eval() ，那么我会收到错误You must feed a value for placeholder tensor.

如果我删除了 eval()，那么一切都会按预期正常工作。

num_steps = 3001

with tf.Session(graph=graph) as session:
    tf.global_variables_initializer().run()
    writer = tf.summary.FileWriter("/home/ubuntu/tensorboard", graph=tf.get_default_graph())
    for step in range(num_steps):
        offset = (step * batch_size) % (train_labels.shape[0] - batch_size)
        batch_data = train_dataset[offset:(offset + batch_size), :]
        batch_labels = train_labels[offset:(offset + batch_size), :]
        feed_dict = {tf_train_dataset : batch_data, tf_train_labels : batch_labels}
        _, l, predictions, summary = session.run([optimizer, loss, train_prediction, summary_op], feed_dict=feed_dict)
        writer.add_summary(summary, step)

        # If I removed this line, then it would work
        loss.eval()

batch_size = 128

graph = tf.Graph()
with graph.as_default():
    with tf.name_scope('tf_train_dataset'):
        tf_train_dataset = tf.placeholder(tf.float32, shape=(batch_size, image_size * image_size))
    with tf.name_scope('tf_train_labels'):
        tf_train_labels = tf.placeholder(tf.float32, shape=(batch_size, num_labels))
    with tf.name_scope('tf_valid_dataset'):
        tf_valid_dataset = tf.constant(valid_dataset)
    with tf.name_scope('tf_test_dataset'):
        tf_test_dataset = tf.constant(test_dataset)

    with tf.name_scope('weights'):
        weights = tf.Variable(tf.truncated_normal([image_size * image_size, num_labels]))
    with tf.name_scope('biases'):
        biases = tf.Variable(tf.zeros([num_labels]))

    with tf.name_scope('logits'):
        logits = tf.matmul(tf_train_dataset, weights) + biases
    with tf.name_scope('loss'):
        loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits, tf_train_labels))
        tf.summary.scalar("loss", loss)

    with tf.name_scope('optimizer'):
        optimizer = tf.train.GradientDescentOptimizer(0.5).minimize(loss)

    with tf.name_scope("train_prediction"):
        train_prediction = tf.nn.softmax(logits)
    with tf.name_scope("valid_prediction"):
        valid_prediction = tf.nn.softmax(tf.matmul(tf_valid_dataset, weights) + biases)
    with tf.name_scope("test_prediction"):
        test_prediction = tf.nn.softmax(tf.matmul(tf_test_dataset, weights) + biases)

    with tf.name_scope("correct_prediction"):
        correct_prediction = tf.equal(tf.argmax(tf_train_labels,1), tf.argmax(train_prediction,1))

    with tf.name_scope("accuracy"):
        accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
        tf.summary.scalar("training_accuracy", accuracy)

    summary_op = tf.summary.merge_all()

确切的错误是：

InvalidArgumentError (see above for traceback): You must feed a value for placeholder tensor 'tf_train_dataset/Placeholder' with dtype float and shape [128,784]
     [[Node: tf_train_dataset/Placeholder = Placeholder[dtype=DT_FLOAT, shape=[128,784], _device="/job:localhost/replica:0/task:0/cpu:0"]()]]

有人有更好的方法来记录变量吗？我尝试过tensor_summary，但它没有在网站上显示。

谢谢大家

除了 AllenLavoie 的评论之外，您实际上可以通过 eval 来输入字典。

loss.eval(feed_dict=feed_dict)

TensorFlow 奇怪的 API 不知道我已经事先输入了字典。

Hence: _, l, predictions, summary = session.run([optimizer, loss, train_prediction, summary_op], feed_dict=feed_dict)

即使放置也不起作用before loss.eval()