def DEFAULTPOLICY(state,depth,env): #Rollout simulation for allowed time
logger.debug("DEFAULTPOLICY")
t=depth
reward = state.rew
done = state.terminal()
# RESTOREENV(env, state.envState)
while not done and t < DEPTH_MAX:
a = env.action_space.sample()
if isinstance(a, np.ndarray):
a = a.astype(np.float32)
nextmove = [a]
obs, r, done, info = env.step(nextmove)
reward += r*(0.99**t)
t += 1
if done:
env.reset()
return reward
def get_total_score(self):
return self.game.get_total_reward()
def step(self, action):
self.frame = self.frame + 1
self.receive_action(action)
return self.state, self.reward, self.is_terminal, {}
def close(self):
pass
#---------------------------------------------------------------------------
env = DOOM_ENV()
obs = env.reset()
class QFunction(chainer.Chain):
def __init__(self, n_history=1, n_action=3):
super().__init__(
l1=L.Convolution2D(n_history, 32, ksize=8, stride=4, nobias=False),
l2=L.Convolution2D(32, 64, ksize=3, stride=2, nobias=False),
l3=L.Convolution2D(64, 64, ksize=3, stride=1, nobias=False),
l4=L.Linear(3136, 512),
out=L.Linear(512, n_action, initialW=np.zeros((n_action, 512), dtype=np.float32))
)
def __call__(self, x, test=False):
s = chainer.Variable(x)
h1 = F.relu(self.l1(s))
h2 = F.relu(self.l2(h1))
if args.render:
misc.env_modifiers.make_rendered(env)
def __exit__(self, *args):
pass
env.__exit__ = __exit__
return env
n_episodes = args.n_episodes
n_sims = args.num_sims
env = make_env() #Evaluation environment
max_episode_len = env.spec.tags.get(
'wrapper_config.TimeLimit.max_episode_steps')
obs_size = np.asarray(env.observation_space.shape).prod()
action_space = env.action_space
env.reset()
logger.info('Maximum episode length %s, number of actions %s'%(max_episode_len,action_space.n))
logger.info('Reward scale %s, Tmax %s, Max_depth %s, Number of simulation %s'%(args.reward_scale_factor, T_MAX, DEPTH_MAX, n_sims))
MCTS_env = make_env() #Simulation environment
MCTS_env.reset()
scores = []
prev_time = start_time
for i in range(n_episodes):
env.reset()
state = CLONEENV(env)
# print (state)
done = False
test_r = 0
def step(self, action):
self.frame = self.frame + 1
self.receive_action(action)
return self.state, self.reward, self.is_terminal, {}
def set_visible(self, visible):
self.game.set_window_visible(visible)
def close(self):
pass
#---------------------------------------------------------------------------
env = DOOM_ENV()
obs = env.reset()
class QFunction(chainer.Chain):
def __init__(self, n_history=1, n_action=6):
super().__init__(
l1=L.Convolution2D(n_history, 32, ksize=8, stride=4, nobias=False),
l2=L.Convolution2D(32, 64, ksize=3, stride=2, nobias=False),
l3=L.Convolution2D(64, 64, ksize=3, stride=1, nobias=False),
l4=L.Linear(3136, 512),
out=L.Linear(512, n_action, initialW=np.zeros((n_action, 512), dtype=np.float32))
)
def __call__(self, x, test=False):
s = chainer.Variable(x)
h1 = F.relu(self.l1(s))
h2 = F.relu(self.l2(h1))