Error when computing summaries in TensorFlow

Question

I am trying to use TensorFlow to produce summaries and visualize them using TensorBoard. However, I am getting an error (InvalidArgumentError: You must feed a value for placeholder tensor 'Placeholder_1' with dtype float) that I do not understand.

This is the full source of my program:

from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)

import tensorflow as tf

x = tf.placeholder(tf.float32, [None, 784])
W = tf.Variable(tf.zeros([784, 10]))
b = tf.Variable(tf.zeros([10]))
y = tf.nn.softmax(tf.matmul(x, W) + b)

_ = tf.histogram_summary("weights", W)
_ = tf.histogram_summary("biases", b)
_ = tf.histogram_summary("y", y)


y_ = tf.placeholder(tf.float32, [None, 10])

with tf.name_scope("xent") as scope:
    cross_entropy = -tf.reduce_sum(y_*tf.log(y))
    _ = tf.scalar_summary("cross entropy", cross_entropy)

with tf.name_scope("train") as scope:
    train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)

init = tf.initialize_all_variables()

sess =  tf.Session()
sess.run(init)

with tf.name_scope("test") as scope:
    correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
    _ = tf.scalar_summary("accuracy", accuracy)

merged = tf.merge_all_summaries()
writer = tf.train.SummaryWriter("/tmp/mnist_nn", sess.graph_def)

for i in range(1000):
    if (i % 10) == 0:
        feed = {x: mnist.test.images, y_: mnist.test.labels}
        result = sess.run([merged, accuracy], feed_dict=feed) 
        summary_str = result[0]
        acc = result[1]
        print("Accuracy at step %s: %s" % (i, acc))
    else:
        batch_xs, batch_ys = mnist.train.next_batch(100)
        feed = {x: batch_xs, y_: batch_ys}
        sess.run(train_step, feed_dict=feed)

print(accuracy.eval({x: mnist.test.images, y_: mnist.test.labels}))

However, when I attempt to run the above code, the following error is raised:

---------------------------------------------------------------------------
InvalidArgumentError                      Traceback (most recent call last)
<ipython-input-23-584a7bc91816> in <module>()
     39     if (i % 10) == 0:
     40         feed = {x: mnist.test.images, y_: mnist.test.labels}
---> 41         result = sess.run([merged, accuracy], feed_dict=feed)
     42         summary_str = result[0]
     43         acc = result[1]

/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/tensorflow/python/client/session.pyc in run(self, fetches, feed_dict)
    366 
    367     # Run request and get response.
--> 368     results = self._do_run(target_list, unique_fetch_targets, feed_dict_string)
    369 
    370     # User may have fetched the same tensor multiple times, but we

/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/tensorflow/python/client/session.pyc in _do_run(self, target_list, fetch_list, feed_dict)
    442         # pylint: disable=protected-access
    443         raise errors._make_specific_exception(node_def, op, error_message,
--> 444                                               e.code)
    445         # pylint: enable=protected-access
    446       six.reraise(e_type, e_value, e_traceback)

InvalidArgumentError: You must feed a value for placeholder tensor 'Placeholder_1' with dtype float
     [[Node: Placeholder_1 = Placeholder[dtype=DT_FLOAT, shape=[], _device="/job:localhost/replica:0/task:0/cpu:0"]()]]
Caused by op u'Placeholder_1', defined at:

It seems from the error that a placeholder in my source has not been fed an appropriate value. As far as I can tell, I am feeding values for all of the placeholders (x and y_).

If you need I'll add the full log to this issue.

I also found that when I first fetch mnist, it does work (with the following output) but still no TensorBoard visualization is produced:

Successfully downloaded train-images-idx3-ubyte.gz 9912422 bytes.
Extracting MNIST_data/train-images-idx3-ubyte.gz
Successfully downloaded train-labels-idx1-ubyte.gz 28881 bytes.
Extracting MNIST_data/train-labels-idx1-ubyte.gz
Successfully downloaded t10k-images-idx3-ubyte.gz 1648877 bytes.
Extracting MNIST_data/t10k-images-idx3-ubyte.gz
Successfully downloaded t10k-labels-idx1-ubyte.gz 4542 bytes.
Extracting MNIST_data/t10k-labels-idx1-ubyte.gz
Tensor("MergeSummary/MergeSummary:0", shape=TensorShape([]), dtype=string)
merged
Accuracy at step 0: 0.098
Accuracy at step 10: 0.7404
Accuracy at step 20: 0.8041
Accuracy at step 30: 0.814 ...

Answer 1

From your error message, it looks like you are using IPython. One pitfall when using IPython to build a TensorFlow model is that functions like tf.merge_all_summaries() will remember every summary created in the current session, including cells that failed with an error. This is a result of TensorFlow using a default graph to collect all of the operations, summaries, etc. that are created in a process, unless you specify the graph explicitly. I suspect that your call to tf.merge_all_summaries() is returning more than the three histogram summaries that you created in your code, and the older ones will have a dependency on a previously created placeholder.

There are two main ways that you could fix that. The simplest is to explicitly merge the summaries, rather than using tf.merge_all_summaries():

weights_summary = tf.histogram_summary("weights", W)
biases_summary = tf.histogram_summary("biases", b)
y_summary = tf.histogram_summary("y", y)

merged = tf.merge_summary([weights_summary, biases_summary, y_summary])

An alternative would be to set an explicit default graph before constructing your model. This is awkward if you want to split your model across multiple IPython cells, but should also work:

# Sets a new default graph, and stores it in `g`.
with tf.Graph().as_default() as g:  

    x = tf.placeholder(tf.float32, [None, 784])
    # ...

    # Summaries are added to `g`.
    _ = tf.histogram_summary("weights", W)
    _ = tf.histogram_summary("biases", b)
    _ = tf.histogram_summary("y", y)

    # ...

    # `merged` contains only summaries from `g`.
    merged = tf.merge_all_summaries()

    # ...

Answer 2

for keras users

you might encounter this error when using the TensorBoard callback and fitting a new model from scratch. In this case, the solution is to call

from keras import backend as K
K.clear_session()

before creating the new model. verified with keras 2.1.5 and tensorflow 1.6.0