Finally, after all the digression on reinforcement learning and DQN and writing down the complete code for the project, you can run it using a script or a Jupyter Notebook, leveraging the Environment class that puts all the code functionalities together:
lunar_lander = Environment(game="LunarLander-v2")
After instantiating it, you just have to run the train, starting from epsilon=1.0 and setting the goal to 5000 episodes (which corresponds to about 2.2 million examples of chained variables of state, action and reward). The actual code we provided is set to successfully accomplish a fully trained DQN model, though it may take some time, given your GPU's availability and its computing capabilities: ...
