def random_policy_test(steps=100, verbose=False):
print("==== Results with Random Policy ====")
config = Config()
actions = 3
config.current_step = 0
env = MountainCar(config)
cumulative_reward = 0
terminations = 0
steps_per_episode = []
episode_steps = 0
for i in range(steps):
A = np.random.randint(actions)
old_state = env.get_current_state()
next_S, R, terminate = env.step(A)
if verbose:
print("Old state:", np.round(old_state, 3), "-->", "Action:", A,
"-->", "New state:", np.round(next_S, 3))
cumulative_reward += R
episode_steps += 1
if terminate:
if verbose:
print("\n## Reset ##\n")
if terminate:
terminations += 1
steps_per_episode.append(episode_steps)
episode_steps *= 0
env.reset()
if not terminate:
steps_per_episode.append(episode_steps)
print("Number of steps per episode:", steps_per_episode)
print("Number of episodes that reached the end:", terminations)
average_length = np.average(episode_steps)
print("The average number of steps per episode was:", average_length)
print("Cumulative reward:", cumulative_reward)
print("\n\n")
def pumping_action_test(steps=100, verbose=False):
print("==== Results with Pumping Action Policy ====")
config = Config()
config.current_step = 0
env = MountainCar(config)
steps_per_episode = []
return_per_episode = []
episode_steps = 0
episode_return = 0
terminations = 0
for i in range(steps):
current_state = env.get_current_state()
A = 1 + np.sign(current_state[1])
old_state = env.get_current_state()
next_S, R, terminate = env.step(A)
if verbose:
print("Old state:", np.round(old_state, 3), "-->", "Action:", A,
"-->", "New state:", np.round(next_S, 3))
episode_steps += 1
episode_return += R
if terminate:
terminations += 1
if verbose:
print("\n## Reset ##\n")
env.reset()
steps_per_episode.append(episode_steps)
return_per_episode.append(episode_return)
episode_steps *= 0
episode_return *= 0
print("Number of steps per episode:", steps_per_episode)
print("Number of successful episodes:", terminations)
print("Return per episode:", return_per_episode)
print("The average return per episode is:", np.mean(return_per_episode))