class BaseEnv(gym.Env):
metadata = {'render.modes': ['human', 'rgb_array']}
def __init__(self, display_time=50, **kwargs):
self.game_name = 'Game Name'
self.display_time = display_time
self.init(**kwargs)
def init(self, **kwargs):
self.game = Environment(env_name=self.game_name, **kwargs)
self.action_set = self.game.env.action_map
self.action_space = spaces.Discrete(self.game.num_actions())
self.observation_space = spaces.Box(0.0,
1.0,
shape=self.game.state_shape(),
dtype=np.float32)
def step(self, action):
reward, done = self.game.act(action)
return (self.game.state(), reward, done, {})
def reset(self):
self.game.reset()
return self.game.state()
def seed(self, seed=None):
self.game = Environment(env_name=self.game_name, random_seed=seed)
return seed
def render(self, mode='human'):
if mode == 'rgb_array':
return self.game.state()
elif mode == 'human':
self.game.display_state(self.display_time)
def close(self):
if self.game.visualized:
self.game.close_display()
return 0
G = 0.0
# Initialize the environment
env.reset()
terminated = False
# Obtain first state, unused by random agent, but included for illustration
s = env.state()
while (not terminated):
# Select an action uniformly at random
action = random.randrange(num_actions)
# Act according to the action and observe the transition and reward
reward, terminated = env.act(action)
# Obtain s_prime, unused by random agent, but included for illustration
s_prime = env.state()
env.display_state(50)
G += reward
# Increment the episodes
e += 1
# Store the return for each episode
returns.append(G)
print("Avg Return: " + str(numpy.mean(returns)) + "+/-" +
str(numpy.std(returns) / numpy.sqrt(NUM_EPISODES)))