How to use a tensorflow session inside a nerual network model

Question

I was following along with a sentdex tutorial on writing a constitutional neural network and I got to wondering if I could figure out my own pooling layer. The problem is that as part of this pooling layer I have to perform a tensorflow function using a session.

def customPool(x):
patches = tf.extract_image_patches(x, [1, 2, 2, 1], [1,2,2,1], [1,1,1,1], 'SAME')

tempSess = tf.Session()
bool1 = tempSess.run( tf.greater( tf.reduce_max(patches) , tf.contrib.distributions.percentile(patches, q=75.) ) )
tempSess.close()

if bool1:
    return tf.nn.max_pool(x, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')
else:
    return tf.nn.avg_pool(x, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')

but the problem is, at least I think, that since I start everything out with a place holder

x = tf.placeholder('float', [None, 784])

Basically my question is: How do I compute something using a tensorflow session inside of the neural network model if it is being passed placeholder variables? Help is much appreciated! full code:

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("/tmp/data/", one_hot = True)

n_classes = 10
batch_size = 128

x = tf.placeholder('float', [None, 784])
y = tf.placeholder('float')

keep_rate = 0.8
keep_prob = tf.placeholder(tf.float32)


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

def maxpool2d(x):
    #                        size of window         movement of window
    return tf.nn.max_pool(x, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')

def customPool(x):
    patches = tf.extract_image_patches(x, [1, 2, 2, 1], [1,2,2,1], [1,1,1,1], 'SAME')

    tempSess = tf.Session()
    bool1 = tempSess.run( tf.greater( tf.reduce_max(patches) , tf.contrib.distributions.percentile(patches, q=75.) ) )
    tempSess.close()

    if bool1:
        return tf.nn.max_pool(x, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')
    else:
        return tf.nn.avg_pool(x, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')

def convolutional_neural_network(x):
    weights = {'W_conv1':tf.Variable(tf.random_normal([5,5,1,32])),
           'W_conv2':tf.Variable(tf.random_normal([5,5,32,64])),
           'W_fc':tf.Variable(tf.random_normal([7*7*64,1024])),
           'out':tf.Variable(tf.random_normal([1024, n_classes]))}

    biases = {'b_conv1':tf.Variable(tf.random_normal([32])),
           'b_conv2':tf.Variable(tf.random_normal([64])),
           'b_fc':tf.Variable(tf.random_normal([1024])),
           'out':tf.Variable(tf.random_normal([n_classes]))}

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

    conv1 = tf.nn.relu(conv2d(x, weights['W_conv1']) + biases['b_conv1'])
    #conv1 = maxpool2d(conv1)
    conv1 = customPool(conv1)

    conv2 = tf.nn.relu(conv2d(conv1, weights['W_conv2']) + biases['b_conv2'])
    #conv2 = maxpool2d(conv2)
    conv1 = customPool(conv1)

    fc = tf.reshape(conv2,[-1, 7*7*64])
    fc = tf.nn.relu(tf.matmul(fc, weights['W_fc'])+biases['b_fc'])
    fc = tf.nn.dropout(fc, keep_rate)

    output = tf.matmul(fc, weights['out'])+biases['out']

    return output

def train_neural_network(x):
    prediction = convolutional_neural_network(x)
    cost = tf.reduce_mean( tf.nn.softmax_cross_entropy_with_logits(logits=prediction, labels=y) )
    optimizer = tf.train.AdamOptimizer().minimize(cost)


    hm_epochs = 1
    with tf.Session() as sess:
        sess.run(tf.initialize_all_variables())

        saver = tf.train.Saver()

        for epoch in range(hm_epochs):
            epoch_loss = 0
            for _ in range(int(mnist.train.num_examples/batch_size)):
            epoch_x, epoch_y = mnist.train.next_batch(batch_size)
            _, c = sess.run([optimizer, cost], feed_dict={x: epoch_x, y: epoch_y})
                epoch_loss += c

    print('Epoch', epoch, 'completed out of',hm_epochs,'loss:',epoch_loss)

    correct = tf.equal(tf.argmax(prediction, 1), tf.argmax(y, 1))

    accuracy = tf.reduce_mean(tf.cast(correct, 'float'))
    print('Accuracy:',accuracy.eval({x:mnist.test.images, y:mnist.test.labels}))

    save_path = saver.save(sess, "/tmp/convnet_maxpool")
    print("Model saved in file: %s" % save_path)


#sess = tf.Session()
train_neural_network(x)
#sess.close()

EDIT: after following Maxim's advice I ran it and it threw the error

_, c = sess.run([optimizer, cost], feed_dict={x: epoch_x, y: epoch_y})
InvalidArgumentError (see above for traceback): logits and labels must be same size: logits_size=[512,10] labels_size=[128,10]
     [[Node: SoftmaxCrossEntropyWithLogits = SoftmaxCrossEntropyWithLogits[T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"](Reshape_2, Reshape_3)]]

it was tracing back to:

File "conv net custom test 1.py", line 89, in <module>
   train_neural_network(x)
File "conv net custom test 1.py", line 59, in train_neural_network
   cost=tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=prediction, labels=y) )

Answer 1

Use tf.cond, not session:

def customPool(x):
  patches = tf.extract_image_patches(x, [1, 2, 2, 1], [1,2,2,1], [1,1,1,1], 'SAME')
  pred = tf.greater(tf.reduce_max(patches), 
                    tf.contrib.distributions.percentile(patches, q=75.))
  return tf.cond(pred, 
                 lambda: tf.nn.max_pool(x, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME'),
                 lambda: tf.nn.avg_pool(x, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME'))

Update:

You've also a copy-paste bug: it's conv1 = customPool(conv1) twice in a row, conv2 isn't downsampled, hence the dimensions error.