As you may have noticed, our agent has several hyperparameters like the learning rate, gamma, epsilon start/minimum value, and so on. There are also several configuration parameters for both the agent and the environment that we would want to be able to modify easily and run instead of searching through the code to find where that parameter was defined. Having a simple and good way to manage these parameters also helps when we want to automate the training process or run parameter sweeps or other methods to tune and find the best set of parameters that work for the agent.

In the following two subsections, we will look at how we can use a JSON file to specify the parameters and hyperparameters in an easy to use way and implement a parameter manager class to handle these externally configurable parameters to update the agent and the environment configuration.