We have successfully coded an agent to use a neural network deep learning architecture to solve a problem. Let's now look at a few ways that we could improve our model. Unlike other machine learning, we cannot evaluate to a performance metric as usual, where we try to minimize some chosen error rate. Success in reinforcement learning is slightly more subjective. You may want an agent to complete a task as quickly as possible, to acquire as many points as possible, or to make the fewest mistakes possible. In addition, depending on the task, we may be able to alter the agent itself to see how it impacts results.
We will look at three possible methods for improving performance:
- Action size: At times, this will be an option and, at times, it will not. If you are trying to solve a problem where the agent rules and environment rules...
