def create_doom(env_id,
client_id,
envWrap=True,
record=False,
outdir=None,
noLifeReward=False,
acRepeat=0,
**_):
from ppaquette_gym_doom import wrappers
if 'labyrinth' in env_id.lower():
if 'single' in env_id.lower():
env_id = 'ppaquette/LabyrinthSingle-v0'
elif 'fix' in env_id.lower():
env_id = 'ppaquette/LabyrinthManyFixed-v0'
else:
env_id = 'ppaquette/LabyrinthMany-v0'
elif 'very' in env_id.lower():
env_id = 'ppaquette/DoomMyWayHomeFixed15-v0'
elif 'sparse' in env_id.lower():
env_id = 'ppaquette/DoomMyWayHomeFixed-v0'
elif 'fix' in env_id.lower():
if '1' in env_id or '2' in env_id:
env_id = 'ppaquette/DoomMyWayHomeFixed' + str(env_id[-2:]) + '-v0'
elif 'new' in env_id.lower():
env_id = 'ppaquette/DoomMyWayHomeFixedNew-v0'
else:
env_id = 'ppaquette/DoomMyWayHomeFixed-v0'
else:
env_id = 'ppaquette/DoomMyWayHome-v0'
# VizDoom workaround: Simultaneously launching multiple vizdoom processes
# makes program stuck, so use the global lock in multi-threading/processing
client_id = int(client_id)
time.sleep(client_id * 10)
env = gym.make(env_id)
modewrapper = wrappers.SetPlayingMode('algo')
obwrapper = wrappers.SetResolution('160x120')
acwrapper = wrappers.ToDiscrete('minimal')
env = modewrapper(obwrapper(acwrapper(env)))
# env = env_wrapper.MakeEnvDynamic(env) # to add stochasticity
if record and outdir is not None:
env = gym.wrappers.Monitor(env, outdir, force=True)
if envWrap:
fshape = (42, 42)
frame_skip = acRepeat if acRepeat > 0 else 4
env.seed(None)
if noLifeReward:
env = env_wrapper.NoNegativeRewardEnv(env)
env = env_wrapper.BufferedObsEnv(env, skip=frame_skip, shape=fshape)
env = env_wrapper.SkipEnv(env, skip=frame_skip)
elif noLifeReward:
env = env_wrapper.NoNegativeRewardEnv(env)
env = Vectorize(env)
env = DiagnosticsInfo(env)
env = Unvectorize(env)
return env
def __init__(self, level='ppaquette/DoomMyWayHome-v0'):
time.sleep(np.random.randint(100))
env = gym.make(level)
modewrapper = wrappers.SetPlayingMode('algo')
obwrapper = wrappers.SetResolution('160x120')
acwrapper = wrappers.ToDiscrete('minimal')
env = modewrapper(obwrapper(acwrapper(env)))
frame_skip = options.get('environment/frame_skip', None)
if frame_skip is not None:
skip_wrapper = SkipWrapper(frame_skip)
env = skip_wrapper(env)
self._record = options.get('environment/record', False)
if self._record:
out_dir = options.get('environment/out_dir',
'/tmp/' + level.split('/')[-1])
if not os.path.exists(out_dir):
os.makedirs(out_dir)
env = gym.wrappers.Monitor(env, out_dir, force=True)
self._no_op_max = options.get('environment/no_op_max', 0)
self._reset_action = env.action_space.sample() \
if options.get('environment/stochastic_reset', False) else 0
env.seed(random.randrange(1000000))
self._show_ui = options.get('show_ui', False)
limit = options.get(
'environment/limit',
env.spec.tags.get('wrapper_config.TimeLimit.max_episode_steps'))
if limit is not None:
env._max_episode_steps = limit
shape = options.get('environment/shape', (42, 42))
self._shape = shape[:2]
self._channels = 0 if len(shape) == 2 else shape[-1]
self.action_size = self._get_action_size(env)
if self.action_size != options.algorithm.output.action_size:
print(
'Algorithm expects different action size (%d) from gym (%d). \n'
'Please set correct action size in you configuration yaml.' %
(options.algorithm.output.action_size, self.action_size))
sys.exit(-1)
self.env = NoNegativeRewardEnv(env)
self._obs_buffer = deque(maxlen=2)
self.observation_space = Box(0.0, 255.0, shape)
self.observation_space.high[...] = 1.0
self._scale = (1.0 / 255.0)
self.reset()
def create_doom(record=False, outdir=None):
from ppaquette_gym_doom import wrappers
env = gym.make('ppaquette/DoomMyWayHome-v0')
modewrapper = wrappers.SetPlayingMode('algo')
obwrapper = wrappers.SetResolution('160x120')
acwrapper = wrappers.ToDiscrete('minimal')
env = modewrapper(obwrapper(acwrapper(env)))
if record:
env = gym.wrappers.Monitor(env, outdir, force=True)
fshape = (42, 42)
env.seed(None)
#env = env_wrapper.NoNegativeRewardEnv(env)
env = BufferedObsEnv(env, skip=1, shape=fshape)
return env