Tabular Q-learning on OpenAI Gym's Frozen Lake. Samples from the observation space, updating the Q-value of each state/action pair. Starts by exploring the observation space through taking random actions, then over time exploits the known Q-values by taking the argmax at the current state.