Artificial Intelligence By Example

Training

As in Chapter 9, Getting Your Neurons to Work, an algorithm with adaptive learning rates trains the model. The Adam algorithm using a cross-entropy loss function just like Chapter 9, Getting Your Neurons to Work, as shown here:

with tf.name_scope('train'):train_step = tf.train.AdamOptimizer(FLAGS.learning_rate).minimize(cross_entropy)

The similarity between the Keras model in Chapter 9, Getting Your Neurons to Work, and this TensorFlow model does not come as a surprise since Keras uses TensorFlow as a backend.

The train node has several inputs from other nodes for its Adam optimizer—cross-entropy, layer2, dropout, layer1, and x-input, as shown in the following graph:

Train node

At this point, TensorFlow provides all the necessary ...