Hands-On Artificial Intelligence for Beginners

Training methods

First, let's define our training method. We'll call this function trainingPipeline; it will take in an action input as well as a y input which represents the targeting Q-value, which we'll define here as placeholders, and calculate a Q-value for an action based on those action/state pairs: 

We'll use a Mean Squared Error (MSE) loss function and calculate it, utilizing the predicted Q-value minus the actual Q-value. Lastly, you might notice that we are using an optimizer here that you might not be familiar with, RMSProp. It's an adaptive learning rate optimizer similar to Adam that was proposed by Geoffrey Hinton. We won't ...