Placeholder missing error in Tensor flow for CNN

Question

I am using tensor flow to run a convolution neural network on MNIST database. But I am getting the following error.

tensorflow.python.framework.errors.InvalidArgumentError: You must feed a value for placeholder tensor 'x' with dtype float [[Node: x = Placeholderdtype=DT_FLOAT, shape=[], _device="/job:localhost/replica:0/task:0/cpu:0"]]

x = tf.placeholder(tf.float32, [None, 784], name='x') # mnist data image of shape 28*28=784

I thought I correctly update the value of x using feed_dict, but its saying i haven't update the value of placeholder x.

Also, is there any other logical flaw in my code?

Any help would be greatly appreciated. Thanks.

import tensorflow as tf
import numpy
from tensorflow.examples.tutorials.mnist import input_data

def conv2d(x, W):
  return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')

def max_pool_2x2(x):
  return tf.nn.max_pool(x, ksize=[1, 2, 2, 1],
                        strides=[1, 2, 2, 1], padding='SAME')

def weight_variable(shape):
  initial = tf.truncated_normal(shape, stddev=0.1)
  return tf.Variable(initial)

def bias_variable(shape):
  initial = tf.constant(0.1, shape=shape)
  return tf.Variable(initial)


mnist = input_data.read_data_sets("/tmp/data/", one_hot=True)

# Parameters
learning_rate = 0.01
training_epochs = 10
batch_size = 100
display_step = 1

# tf Graph Input
#x = tf.placeholder(tf.float32, [50, 784], name='x') # mnist data image of shape 28*28=784
#y = tf.placeholder(tf.float32, [50, 10], name='y') # 0-9 digits recognition => 10 classes

# Set model weights
W = tf.Variable(tf.zeros([784, 10]), name="weights")
b = tf.Variable(tf.zeros([10]), name="bias")

W_conv1 = weight_variable([5, 5, 1, 32])
b_conv1 = bias_variable([32])


W_conv2 = weight_variable([5, 5, 32, 64])
b_conv2 = bias_variable([64])


W_fc1 = weight_variable([7 * 7 * 64, 1024])
b_fc1 = bias_variable([1024])

W_fc2 = weight_variable([1024, 10])
b_fc2 = bias_variable([10])

# Initializing the variables
init = tf.initialize_all_variables()

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


    # Training cycle
    for i in range(1000):
        print i
        batch_xs, batch_ys = mnist.train.next_batch(50)

        x_image = tf.reshape(x, [-1,28,28,1])

        h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
        h_pool1 = max_pool_2x2(h_conv1)

        h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
        h_pool2 = max_pool_2x2(h_conv2)

        h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])
        h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)


        y_conv=tf.nn.softmax(tf.matmul(h_fc1, W_fc2) + b_fc2)

        cross_entropy = tf.reduce_mean(-tf.reduce_sum(y * tf.log(y_conv), reduction_indices=[1]))
        sess.run(
          [cross_entropy, y_conv],
          feed_dict={x: batch_xs, y: batch_ys})

        correct_prediction = tf.equal(tf.argmax(y_conv,1), tf.argmax(y,1))
        print correct_prediction.eval()
        accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

Answer 1

Why are you trying to create placeholder variables ? You should be able to use the outputs generated by mnist.train.next_batch(50) directly provided that you move the computation of correct_prediction and accuracy inside the model itself.

batch_xs, batch_ys = mnist.train.next_batch(50)
x_image = tf.reshape(batch_xs, [-1,28,28,1])
...
cross_entropy = tf.reduce_mean(-tf.reduce_sum(batch_ys * tf.log(y_conv), reduction_indices=[1]))
correct_prediction = tf.equal(tf.argmax(y_conv,1), tf.argmax(batch_ys,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
predictions_correct, acc = sess.run([cross_entropy, y_conv, correct_prediction, accuracy])
print predictions_correct, acc

Answer 2

You're receiving that error because you're attempting to run eval() on correct_prediction . That tensor requires the batch inputs (x and y) in order to be evaluated. You could correct the error by changing it to:

print correct_prediction.eval(feed_dict={x: batch_xs, y: batch_ys})

But as Benoit Steiner mentioned, you could just as easily pull it into the model.

On a more general note, you're not doing any kind of optimization here, but maybe you just haven't gotten around to that yet. As it stands now, it'll just print out bad predictions for a while. :)

Answer 3

Firstly your x and y are commented out, if this is present in your actual code it is very likely the issue.

correct_prediction.eval() is equivalent to tf.session.run(correct_prediction) (or in your case sess.run() ) and thus requires the same syntax*. So it would need to be correct_prediction.eval(feed_dict={x: batch_xs, y: batch_ys}) in order to run, be warned however that this is generally RAM intensive, and may cause your system to hang. Pulling the accuracy function into the model may be a good idea because of the ram usage.

I did not see an optimization function to utilize your cross entropy, however i have never tried not using one,so if it works don't fix it. but if it ends up throwing an error you may want to try:

optimizer = optimizer = tf.train.AdamOptimizer().minimize(cross_entropy)

and replace the ' cross_entropy ' in

sess.run([cross_entropy, y_conv],feed_dict={x: batch_xs, y: batch_ys})

with ' optimizer '

https://pythonprogramming.net/tensorflow-neural-network-session-machine-learning-tutorial/

check the accuracy evaluation section of the script.