Hands-On Q-Learning with Python

Greedy strategy

The greedy strategy tells us we should always pick the highest-valued action we have seen so far, without doing any exploration at all. As we've discussed, this can lead to getting stuck on a local maximum, or an action that currently has a higher Q-value than the other actions we could take from that state.

Using the greedy strategy, we would continue to follow that local maximum reward path even though there might be other potential actions we could take that would have higher Q-values once we progressed through those reward gradients. For that reason, we see that the greedy strategy would allow us to get stuck on a suboptimal action path forever, with no option for exploring alternatives.