Python: Advanced Guide to Artificial Intelligence

Exploration and exploitation in the RL algorithms

In the absence of a model, at every step the agent either explores or exploits. Exploration means that the agent selects an unknown action to find out the reward and the model. Exploitation means that the agent selects the best-known action to get the maximum reward. If the agent always decides to exploit then it might get stuck in a local optimal value. Hence, sometimes the agent takes a detour from learned policy to explore unknown actions. Similarly, if an agent always decides to explore then it may fail to find an optimal policy. Thus, it is important to have a balance of exploration and exploitation. In our code, we implement this by using a probability p to select a random action and ...