We now have a learning agent that uses a neural network to learn Q-values and update itself to perform better at the task. The agent takes a while to learn before it starts acting wisely. How do we know what is going on with the agent at a given time? How do we know if the agent is making progress or simply acting dumb? How do we see and measure the progress of the agent with time? Should we just sit and wait for the training to end? No. There should be some better way, don't you think?
Yes, and there is! It is actually important for us, the developers of the agents, to be able to observe how the agent is performing in order to figure out if there is an issue with the implementation or if ...