Hands-On Reinforcement Learning for Games

Playing the FrozenLake game

The agent code now plays or explores the environment and it is helpful if we understand how this code runs. Open up Chapter_3_3.py again and follow the exercise:

1. All we need to focus on for this section is how the agent plays the game. Scroll down to the play_game function, as shown in the following:

def play_game(env, policy, display=True):  env.reset()  episode = []  finished = False  while not finished:    s = env.env.s    if display:      clear_output(True)      env.render()      sleep(1)    timestep = []    timestep.append(s)    n = random.uniform(0, sum(policy[s].values()))    top_range = 0    action = 0    for prob in policy[s].items():      top_range += prob[1]                  if n < top_range:        action = prob[0]        break  state, reward, finished, info = env.step(action) ...