Skip to content
Branch: master
Find file Copy path
Find file Copy path
1 contributor

Users who have contributed to this file

292 lines (239 sloc) 9.27 KB
import numpy as np
from collections import deque
import gym
from gym import spaces
import cv2
def is_atari(env):
if (hasattr(env.observation_space, "shape")
and env.observation_space.shape is not None
and len(env.observation_space.shape) <= 2):
return False
return hasattr(env, "unwrapped") and hasattr(env.unwrapped, "ale")
def get_wrapper_by_cls(env, cls):
"""Returns the gym env wrapper of the given class, or None."""
currentenv = env
while True:
if isinstance(currentenv, cls):
return currentenv
elif isinstance(currentenv, gym.Wrapper):
currentenv = currentenv.env
return None
class MonitorEnv(gym.Wrapper):
def __init__(self, env=None):
"""Record episodes stats prior to EpisodicLifeEnv, etc."""
gym.Wrapper.__init__(self, env)
self._current_reward = None
self._num_steps = None
self._total_steps = None
self._episode_rewards = []
self._episode_lengths = []
self._num_episodes = 0
self._num_returned = 0
def reset(self, **kwargs):
obs = self.env.reset(**kwargs)
if self._total_steps is None:
self._total_steps = sum(self._episode_lengths)
if self._current_reward is not None:
self._num_episodes += 1
self._current_reward = 0
self._num_steps = 0
return obs
def step(self, action):
obs, rew, done, info = self.env.step(action)
self._current_reward += rew
self._num_steps += 1
self._total_steps += 1
return (obs, rew, done, info)
def get_episode_rewards(self):
return self._episode_rewards
def get_episode_lengths(self):
return self._episode_lengths
def get_total_steps(self):
return self._total_steps
def next_episode_results(self):
for i in range(self._num_returned, len(self._episode_rewards)):
yield (self._episode_rewards[i], self._episode_lengths[i])
self._num_returned = len(self._episode_rewards)
class NoopResetEnv(gym.Wrapper):
def __init__(self, env, noop_max=30):
"""Sample initial states by taking random number of no-ops on reset.
No-op is assumed to be action 0.
gym.Wrapper.__init__(self, env)
self.noop_max = noop_max
self.override_num_noops = None
self.noop_action = 0
assert env.unwrapped.get_action_meanings()[0] == "NOOP"
def reset(self, **kwargs):
""" Do no-op action for a number of steps in [1, noop_max]."""
if self.override_num_noops is not None:
noops = self.override_num_noops
noops = self.unwrapped.np_random.randint(1, self.noop_max + 1)
assert noops > 0
obs = None
for _ in range(noops):
obs, _, done, _ = self.env.step(self.noop_action)
if done:
obs = self.env.reset(**kwargs)
return obs
def step(self, ac):
return self.env.step(ac)
class ClipRewardEnv(gym.RewardWrapper):
def __init__(self, env):
gym.RewardWrapper.__init__(self, env)
def reward(self, reward):
"""Bin reward to {+1, 0, -1} by its sign."""
return np.sign(reward)
class FireResetEnv(gym.Wrapper):
def __init__(self, env):
"""Take action on reset.
For environments that are fixed until firing."""
gym.Wrapper.__init__(self, env)
assert env.unwrapped.get_action_meanings()[1] == "FIRE"
assert len(env.unwrapped.get_action_meanings()) >= 3
def reset(self, **kwargs):
obs, _, done, _ = self.env.step(1)
if done:
obs, _, done, _ = self.env.step(2)
if done:
return obs
def step(self, ac):
return self.env.step(ac)
class EpisodicLifeEnv(gym.Wrapper):
def __init__(self, env):
"""Make end-of-life == end-of-episode, but only reset on true game over.
Done by DeepMind for the DQN and co. since it helps value estimation.
gym.Wrapper.__init__(self, env)
self.lives = 0
self.was_real_done = True
def step(self, action):
obs, reward, done, info = self.env.step(action)
self.was_real_done = done
# check current lives, make loss of life terminal,
# then update lives to handle bonus lives
lives = self.env.unwrapped.ale.lives()
if lives < self.lives and lives > 0:
# for Qbert sometimes we stay in lives == 0 condtion for a few fr
# so its important to keep lives > 0, so that we only reset once
# the environment advertises done.
done = True
self.lives = lives
return obs, reward, done, info
def reset(self, **kwargs):
"""Reset only when lives are exhausted.
This way all states are still reachable even though lives are episodic,
and the learner need not know about any of this behind-the-scenes.
if self.was_real_done:
obs = self.env.reset(**kwargs)
# no-op step to advance from terminal/lost life state
obs, _, _, _ = self.env.step(0)
self.lives = self.env.unwrapped.ale.lives()
return obs
class MaxAndSkipEnv(gym.Wrapper):
def __init__(self, env, skip=4):
"""Return only every `skip`-th frame"""
gym.Wrapper.__init__(self, env)
# most recent raw observations (for max pooling across time steps)
self._obs_buffer = np.zeros(
(2, ) + env.observation_space.shape, dtype=np.uint8)
self._skip = skip
def step(self, action):
"""Repeat action, sum reward, and max over last observations."""
total_reward = 0.0
done = None
for i in range(self._skip):
obs, reward, done, info = self.env.step(action)
if i == self._skip - 2:
self._obs_buffer[0] = obs
if i == self._skip - 1:
self._obs_buffer[1] = obs
total_reward += reward
if done:
# Note that the observation on the done=True frame
# doesn't matter
max_frame = self._obs_buffer.max(axis=0)
return max_frame, total_reward, done, info
def reset(self, **kwargs):
return self.env.reset(**kwargs)
class WarpFrame(gym.ObservationWrapper):
def __init__(self, env, dim):
"""Warp frames to the specified size (dim x dim)."""
gym.ObservationWrapper.__init__(self, env)
self.width = dim
self.height = dim
self.observation_space = spaces.Box(
shape=(self.height, self.width, 1),
def observation(self, frame):
frame = cv2.cvtColor(frame, cv2.COLOR_RGB2GRAY)
frame = cv2.resize(
frame, (self.width, self.height), interpolation=cv2.INTER_AREA)
return frame[:, :, None]
class FrameStack(gym.Wrapper):
def __init__(self, env, k):
"""Stack k last frames."""
gym.Wrapper.__init__(self, env)
self.k = k
self.frames = deque([], maxlen=k)
shp = env.observation_space.shape
self.observation_space = spaces.Box(
shape=(shp[0], shp[1], shp[2] * k),
def reset(self):
ob = self.env.reset()
for _ in range(self.k):
return self._get_ob()
def step(self, action):
ob, reward, done, info = self.env.step(action)
return self._get_ob(), reward, done, info
def _get_ob(self):
assert len(self.frames) == self.k
return np.concatenate(self.frames, axis=2)
class ScaledFloatFrame(gym.ObservationWrapper):
def __init__(self, env):
gym.ObservationWrapper.__init__(self, env)
self.observation_space = gym.spaces.Box(
low=0, high=1, shape=env.observation_space.shape, dtype=np.float32)
def observation(self, observation):
# careful! This undoes the memory optimization, use
# with smaller replay buffers only.
return np.array(observation).astype(np.float32) / 255.0
def wrap_deepmind(env, dim=84, framestack=True):
"""Configure environment for DeepMind-style Atari.
Note that we assume reward clipping is done outside the wrapper.
dim (int): Dimension to resize observations to (dim x dim).
framestack (bool): Whether to framestack observations.
env = MonitorEnv(env)
env = NoopResetEnv(env, noop_max=30)
if "NoFrameskip" in
env = MaxAndSkipEnv(env, skip=4)
env = EpisodicLifeEnv(env)
if "FIRE" in env.unwrapped.get_action_meanings():
env = FireResetEnv(env)
env = WarpFrame(env, dim)
# env = ScaledFloatFrame(env) # TODO: use for dqn?
# env = ClipRewardEnv(env) # reward clipping is handled by policy eval
if framestack:
env = FrameStack(env, 4)
return env
You can’t perform that action at this time.