handyd

use_batchnorm = True

X = tf.placeholder(tf.float32, [None, input_dim], name = "X")

Y = tf.placeholder(tf.float32, [None, output_dim], name = "Y")

W1 = tf.get_variable("W1", shape=[input_dim, 16], initializer=tf.contrib.layers.xavier_initializer())

b1 = tf.Variable(tf.random_normal([16]))

L1 = tf.matmul(X, W1) + b1;

if use_batchnorm:

    L1 = tf.layers.batch_normalization(L1)

hypothesis = tf.nn.relu(L1)

cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(

        logits=hypothesis, labels=Y))

update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)

with tf.control_dependencies(update_ops):

    optimizer = tf.train.AdamOptimizer(learning_rate = learning_rate).minimize(cost)

...

References:

https://www.tensorflow.org/api_docs/python/tf/layers/batch_normalization

https://github.com/hunkim/DeepLearningZeroToAll/blob/master/lab-10-6-mnist_nn_batchnorm.ipynb

https://shuuki4.wordpress.com/2016/01/13/batch-normalization-%EC%84%A4%EB%AA%85-%EB%B0%8F-%EA%B5%AC%ED%98%84/