Hands-On Artificial Intelligence for Beginners

Forwardprop and backprop with MNIST

Remember that, to run a training process in TensorFlow, we run our computations through a session. To get started, let's open a new training session. 

1. In TensorFlow, before we enter the training process, we need to initialize our model class as an object, and bring all of our variable placeholders online with the tf.global_variables_initializer() command: 

with tf.Session() as sess:    ## Initialize the variable    sess.run(tf.global_variables_initializer())

2. Now, we can write the core of the training process, which is what we call the training loop. We'll define the loop by the number of training epochs for the model. In it, we'll first batch out our incoming data; our model cannot and should not handle all ...