Tensorflow:sess.run（[x]）不工作，但sess.run（[y]）使用相同的feed\u di工作

Question

我正在学习Tensorboard，我正在遵循this tutorial.中的代码

下面是我的代码：

import tensorflow as tf
LOGDIR = "/tmp/mnist_tutorial/"
mnist = tf.contrib.learn.datasets.mnist.read_data_sets(train_dir=LOGDIR + "data", one_hot=True)

def conv_layer(input, size_in, size_out, name="conv"):
    with tf.name_scope(name):
        w = tf.Variable(tf.zeros([5, 5, size_in, size_out]))
        b = tf.Variable(tf.zeros([size_out]))
        conv = tf.nn.conv2d(input, w, strides=[1, 1, 1, 1], padding="SAME")
        act = tf.nn.relu(conv + b)
        tf.summary.histogram("weights", w)
        tf.summary.histogram("biases", b)
        tf.summary.histogram("activations", act)
        return act


def fc_layer(input, size_in, size_out, name="fc"):
    with tf.name_scope(name):
        w = tf.Variable(tf.zeros([size_in, size_out]))
        b = tf.Variable(tf.zeros([size_out]))
        act = tf.nn.relu(tf.matmul(input, w)+b)
        tf.summary.histogram("weights", w)
        tf.summary.histogram("biases", b)
        tf.summary.histogram("activations", act)
        return act

x = tf.placeholder(tf.float32, shape=[None, 784], name='x')
x_image = tf.reshape(x, [-1, 28, 28, 1])
tf.summary.image('input', x_image, 3)

y = tf.placeholder(tf.float32, shape=[None, 10], name='labels')

conv1 = conv_layer(x_image, 1, 32, name='conv1')
pool1 = tf.nn.max_pool(conv1, ksize=[1,2,2,1], strides=[1,2,2,1], padding="SAME")

conv2 = conv_layer(pool1, 32, 64, name='conv2')
pool2 = tf.nn.max_pool(conv2, ksize=[1,2,2,1], strides=[1,2,2,1], padding="SAME")
flattened = tf.reshape(pool2, [-1, 7*7*64])

fc1 = fc_layer(flattened, 7*7*64, 1024, name='fc1')
logits = fc_layer(fc1, 1024, 10, name='fc2')

with tf.name_scope('xent'):
    xent = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=logits, labels=y))
    tf.summary.scalar('cross_entropy', xent)

with tf.name_scope('train'):
    train_step = tf.train.AdamOptimizer(1e-4).minimize(xent)

with tf.name_scope('accruacy'):
    correct_prediction = tf.equal(tf.argmax(logits,1), tf.argmax(y, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
    tf.summary.scalar('accruacy', accuracy)

summ = tf.summary.merge_all()

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    # writer =tf.summary.FileWriter("tmp/mnist_demo/1")
    # writer.add_graph(sess.graph)
    # writer.close()

    for i in range(20):
        batch = mnist.train.next_batch(100)

        # Occasionally report back the accruacy

        if i%2 == 0:
            [train_accruacy] = sess.run([accuracy], feed_dict={x:batch[0], y:batch[1]}) # works
#             [s, train_accruacy] = sess.run([summ, accuracy], feed_dict={x:batch[0], y:batch[1]}) #error!
            print("step %d, training accruacy %g" % (i, train_accruacy))

    sess.run(train_step, feed_dict={x:batch[0],y:batch[1]})

我在使用这行时遇到了一个错误：

[s, train_accruacy] = sess.run([summ, accuracy], feed_dict={x:batch[0], y:batch[1]}) #error!

这是我收到的错误消息：

You must feed a value for placeholder tensor 'x' with dtype float and shape [?,784] [[{{node x}} = Placeholder[dtype=DT_FLOAT, shape=[?,784], _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]

我知道我输入的张量不是（x，784）的正确形状

然而，我不明白为什么[train_accruacy] = sess.run([accuracy], feed_dict={x:batch[0], y:batch[1]}) # works。毕竟，我把相同的东西输入到相同的占位符变量中，它们接受相同形状的张量

除非我完全错了，sess.run([argument], feed_dict=...)的第一个参数描述了要返回的张量。我看不出这会如何影响我输入的数据的形状

另外：这个模型应该有一个错误

对于感兴趣的人，完整的代码是here

是否返回的数据类型也不同tf.summary.merge_all()返回一个字符串张量，但我怀疑这是导致问题的原因

我似乎在网上找不到有关这个问题的任何文件。这应该发生吗

Answer 1

我会回答我自己的问题：

tf.reset_default_graph()工作，添加在def conv_layer()之前

如果不想使用tf.reset_default_graph()：

结果是我在同一个环节中输入了2个张量，这是张量流不允许的

for i in range(20):
    batch = mnist.train.next_batch(100)
    sess.run(train_step, feed_dict={x:batch[0],y:batch[1]})
    # Occasionally report back the accruacy

    if i%2 == 0:
        [train_accruacy] = sess.run([accuracy], feed_dict={x:batch[0], y:batch[1]}) # works
#             [s, train_accruacy] = sess.run([summ, accuracy], feed_dict={x:batch[0], y:batch[1]}) #error!
        print("step %d, training accruacy %g" % (i, train_accruacy))

上面的代码不起作用。结果是，由于某种原因，i%2下面的第一行代码输入张量batch[0]很好，但是，当它注释掉那行并用第二行替换它时，tensorflow似乎没有“清除”出x占位符变量，因此2个独立的张量被输入到输入（来自独立的sess.run()）事件中

此代码适用于：

for i in range(2000):
    batch = mnist.train.next_batch(100)
    if i%10 !=0:
        sess.run(train_step, feed_dict={x:batch[0],y:batch[1]})
        # Occasionally report back the accruacy

    if i%10 == 0:
        [train_accuracy, s] = sess.run([accuracy, summ], feed_dict={x: batch[0], y: batch[1]})
        print("step %d, training accruacy %g" % (i, train_accuracy))
        writer.add_summary(s,i)

这里张量是分开输入的，一切正常

我很高兴如果有人能告诉我为什么会这样，或者这是一个错误