def _create_env_from_random_state(self):
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31.
seed2 = seeding.hash_seed(self.np_random.randint(0, 1000)) % 2**31
seed3 = seeding.hash_seed(seed2 + 1) % 2**31
seed4 = seeding.hash_seed(seed3 + 1) % 2**31
hour = self.np_random.randint(low=0.0, high=24.0)
if self.saved_state is not None:
seed2=self.saved_state[0]
seed3=self.saved_state[1]
seed4=self.saved_state[2]
hour=self.saved_state[3]
print('Using state', seed2, seed3, seed4, hour)
start_time = datetime(2021, 1, 1, hour, 0, 0)
patient = T1DPatient.withName(self.patient_name, random_init_bg=True, seed=seed4)
sensor = CGMSensor.withName(self.SENSOR_HARDWARE, seed=seed2)
if self.harrison_benedict:
vpatient_params = pd.read_csv(PATIENT_PARA_FILE)
self.kind = self.patient_name.split('#')[0]
self.bw = vpatient_params.query('Name=="{}"'.format(self.patient_name))['BW'].item()
scenario=RandomBalancedScenario(start_time=start_time, seed=seed3, harrison_benedict=self.harrison_benedict, bw=self.bw, kind=self.kind)
else:
scenario = RandomScenario(start_time=start_time, seed=seed3)
pump = InsulinPump.withName(self.INSULIN_PUMP_HARDWARE)
env = _T1DSimEnv(patient, sensor, pump, scenario,noise=self.noise)
return env, seed2, seed3, seed4, hour
def _seed(self, seed=None):
self.np_random, seed1 = seeding.np_random(seed=seed)
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31.
seed2 = seeding.hash_seed(seed1 + 1) % 2**31
seed3 = seeding.hash_seed(seed2 + 1) % 2**31
return [seed1, seed2, seed3]
def set_global_seeds(seed):
import tensorflow as tf
from gym.utils.seeding import hash_seed
seed = hash_seed(seed, max_bytes=4)
tf.set_random_seed(seed)
np.random.seed(seed)
random.seed(seed)
def seed(self, seed=None):
self.np_random, seed1 = seeding.np_random(seed)
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31.
seed2 = seeding.hash_seed(seed1 + 1) % 2**31
# Empirically, we need to seed before loading the ROM.
self.ale.setInt(b'random_seed', seed2)
self.ale.loadROM(self.game_path)
if self.game_mode is not None:
modes = self.ale.getAvailableModes()
assert self.game_mode in modes, (
"Invalid game mode \"{}\" for game {}.\nAvailable modes are: {}"
).format(self.game_mode, self.game, modes)
self.ale.setMode(self.game_mode)
if self.game_difficulty is not None:
difficulties = self.ale.getAvailableDifficulties()
assert self.game_difficulty in difficulties, (
"Invalid game difficulty \"{}\" for game {}.\nAvailable difficulties are: {}"
).format(self.game_difficulty, self.game, difficulties)
self.ale.setDifficulty(self.game_difficulty)
return [seed1, seed2]
def _seed(self, seed=None):
self.np_random, seed1 = seeding.np_random(seed)
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31.
seed2 = seeding.hash_seed(seed1 + 1) % 2**31
# Empirically, we need to seed before loading the ROM.
self.ale.setInt(b'random_seed', seed2)
self.ale.loadROM(self.game_path)
self._action_set = self.ale.getMinimalActionSet()
self.action_space = spaces.Discrete(len(self._action_set),
np_random=self.np_random)
(screen_width, screen_height) = self.ale.getScreenDims()
if self._obs_type == 'ram':
self.observation_space = spaces.Box(low=np.zeros(128),
high=np.zeros(128) + 255,
np_random=self.np_random)
elif self._obs_type == 'image':
self.observation_space = spaces.Box(low=0,
high=255,
shape=(screen_height,
screen_width, 3),
np_random=self.np_random)
else:
raise error.Error('Unrecognized observation type: {}'.format(
self._obs_type))
return [seed1, seed2]
def __call__(self, env_maker, seed=None, monitor_file=None):
"""
:param env_maker: instance of roam_learning.robot_env.EnvMaker
:param seed: int that is used to generate seeds for vectorized envs
:param monitor_file: path to a .csv file to log episode rewards, lengths etc,. of the vectorized envs
:return: instance of either DummyVecEnv, SubprocVecEnv or ShmemVecEnv
"""
# Create a list of env makers
if seed is not None:
assert isinstance(seed, int)
env_makers = []
for i in range(self.nenvs):
env_makers += [deepcopy(env_maker)]
if seed is not None:
seed = hash_seed(seed)
env_makers[i].set_seed(seed + i)
# Create the vectorized envs
envs = self.vec_env_wrapper(env_makers)
# Monitor the envs before normalization
if monitor_file is not None:
envs = VecMonitor(envs, filename=monitor_file)
if self.normalize_obs or self.normalize_ret:
envs = VecNormalize(envs, ob=self.normalize_obs, ret=self.normalize_ret, use_tf=True)
return envs
def _create_env_from_random_state(self):
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31.
seed2 = seeding.hash_seed(self.np_random.randint(0, 1000)) % 2**31
seed3 = seeding.hash_seed(seed2 + 1) % 2**31
seed4 = seeding.hash_seed(seed3 + 1) % 2**31
hour = self.np_random.randint(low=0.0, high=24.0)
start_time = datetime(2018, 1, 1, hour, 0, 0)
patient = T1DPatient.withName(self.patient_name, random_init_bg=True, seed=seed4)
sensor = CGMSensor.withName(self.SENSOR_HARDWARE, seed=seed2)
scenario = RandomScenario(start_time=start_time, seed=seed3)
pump = InsulinPump.withName(self.INSULIN_PUMP_HARDWARE)
env = _T1DSimEnv(patient, sensor, pump, scenario)
return env, seed2, seed3, seed4
def seed(self, seed=None):
"""
@brief Specify the seed of the environment.
@details Note that it also resets the low-level Jiminy engine.
One must call the `reset` method manually afterward.
@param[in] seed non-negative integer.
Optional: A strongly random seed will be generated by gym if omitted.
@return Updated seed of the environment
"""
# Generate a 8 bytes (uint64) seed using gym utils
self.rg, self._seed = seeding.np_random(seed)
# Convert it into a 4 bytes uint32 seed.
# Note that hashing is used to get rid off possible
# correlation in the presence of concurrency.
self._seed = np.uint32(
seeding._int_list_from_bigint(seeding.hash_seed(self._seed))[0])
# Reset the seed of Jiminy Engine, if available
if self.engine_py is not None:
self.engine_py.seed(self._seed)
return [self._seed]
def seed(self, seed=None):
if seed is None:
self.seed_ = create_seed()
random.seed(self.seed_)
else:
self.seed_ = hash_seed(seed)
random.seed(self.seed_)
def seed(self, seed=None):
if self.benchmark_mode:
initial_seed = 42
else:
initial_seed = seeding.hash_seed(seed) % 2 ** 32
self._random_state = np.random.RandomState(seed=initial_seed)
return [initial_seed]
def seed(self, seed=None):
"""
Derives a random seed.
"""
self.np_random, seed1 = seeding.np_random(seed)
seed2 = seeding.hash_seed(seed1 + 1) % 2**31
return [seed1, seed2]
def seed(self, seed=None):
print('_seed called')
self.np_random, seed1 = seeding.np_random(seed=seed)
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31.
seed2 = seeding.hash_seed(seed1 + 1) % 2**31
seed3 = seeding.hash_seed(seed2 + 1) % 2**31
hour = self.np_random.randint(low=0.0, high=24.0)
start_time = datetime(2021, 1, 1, hour, 0, 0)
patient = T1DPatient.withName(self.patient_name)
sensor = CGMSensor.withName(self.SENSOR_HARDWARE, seed=seed2)
scenario = RandomScenario(start_time=start_time, seed=seed3)
pump = InsulinPump.withName(self.INSULIN_PUMP_HARDWARE)
self.env = _T1DSimEnv(patient, sensor, pump, scenario)
return [seed1, seed2, seed3]
def seed(self, seed=None):
"""Set seed for the environment random generations."""
self.np_random, seed1 = seeding.np_random(seed)
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31. Takem from atari_env.py
seed2 = seeding.hash_seed(seed1 + 1) % 2**31
self.sim.seed(seed2)
return [seed1, seed2]
def _launch_fceux(self):
# Making sure ROM file is valid
if '' == self.rom_path or not os.path.isfile(self.rom_path):
raise gym.error.Error('Unable to find ROM. Please download the game from the web and configure the rom path by ' +
'calling env.configure(rom_path=path_to_file)')
# Creating pipes
self._create_pipes()
# Creating temporary lua file
self.temp_lua_path = os.path.join('/tmp', str(seeding.hash_seed(None) % 2 ** 32) + '.lua')
temp_lua_file = open(self.temp_lua_path, 'w', 1)
for k, v in list(self.launch_vars.items()):
temp_lua_file.write('%s = "%s";\n' % (k, v))
i = 0
for script in self.lua_path:
temp_lua_file.write('f_%d = assert (loadfile ("%s"));\n' % (i, script))
temp_lua_file.write('f_%d ();\n' % i)
i += 1
temp_lua_file.close()
# Resetting variables
self.last_frame = 0
self.reward = 0
self.episode_reward = 0
self.is_finished = False
self.screen = np.zeros(shape=(self.screen_height, self.screen_width, 3), dtype=np.uint8)
self._reset_info_vars()
# Loading fceux
args = [FCEUX_PATH]
args.extend(self.cmd_args[:])
args.extend(['--loadlua', self.temp_lua_path])
args.append(self.rom_path)
args.extend(['>/dev/null', '2>/dev/null', '&'])
self.subprocess = subprocess.Popen("/bin/bash", shell=False, universal_newlines=True, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
stdout, stderr = self.subprocess.communicate(' '.join(args) + "\necho $!")
self.fceux_pid = int(stdout)
if 0 == self.subprocess.returncode:
self.is_initialized = 1
if not self.disable_out_pipe:
with self.lock_out:
try:
self.pipe_out = open(self.path_pipe_out, 'w', 1)
except IOError:
self.pipe_out = None
# Removing lua file
sleep(1) # Sleeping to make sure fceux has time to load file before removing
if os.path.isfile(self.temp_lua_path):
try:
os.remove(self.temp_lua_path)
except OSError:
pass
else:
self.is_initialized = 0
raise gym.error.Error('Unable to start fceux. Command: %s' % (' '.join(args)))
def _seed(self, seed=None):
np_random, seed1 = seeding.np_random(seed)
# Derive a random seed.
seed2 = seeding.hash_seed(seed1 + 1) % 2**32
self.game.set_seed(seed2)
# 3 allowed actions [12, 13, 14] (must match .cfg file)
self.action_space = spaces.HighLow(np.matrix([[0, 1, 0]] * 3), np_random=np_random)
self.observation_space = spaces.Box(low=0, high=255, shape=(self.screen_height, self.screen_width, 3), np_random=np_random)
return [seed1, seed2]
def _seed(self, seed=None):
np_random, seed1 = seeding.np_random(seed)
# Derive a random seed.
seed2 = seeding.hash_seed(seed1 + 1) % 2**32
self.game.set_seed(seed2)
# action indexes are [0, 9, 10, 12, 13, 14]
self.action_space = spaces.HighLow(np.matrix([[0, 1, 0]] * 6), np_random=np_random)
self.observation_space = spaces.Box(low=0, high=255, shape=(self.screen_height, self.screen_width, 3), np_random=np_random)
return [seed1, seed2]
def _seed(self, seed=None):
self.np_random, seed1 = seeding.np_random(seed)
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31.
seed2 = seeding.hash_seed(seed1 + 1) % 2**31
# Empirically, we need to seed before loading the ROM.
self.ale.setInt(b'random_seed', seed2)
self.ale.loadROM(self.game_path)
return [seed1, seed2]
def seed(self, seed=None):
self.np_random, seed1 = seeding.np_random(seed)
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31.
seed2 = seeding.hash_seed(seed1 + 1) % 2**31
# Empirically, we need to seed before loading the ROM.
self.ale.setInt(b'random_seed', seed2)
self.ale.loadROM(self.game_path)
return [seed1, seed2]
def _seed(self, seed=None):
self.np_random, seed1 = seeding.np_random(seed)
# Derive a random seed.
seed2 = seeding.hash_seed(seed1 + 1) % 2**32
pachi_py.pachi_srand(seed2)
shape = pachi_py.CreateBoard(self.board_size).encode().shape
self.observation_space = spaces.Box(np.zeros(shape), np.ones(shape), np_random=self.np_random)
# One action for each board position, pass, and resign
self.action_space = spaces.Discrete(self.board_size**2 + 2, np_random=self.np_random)
return [seed1, seed2]
def seed(self, seed=None, id=0):
_, seed1 = seeding.np_random(seed)
if id > 0:
seed = seed * 100 + id
self.np_random, _ = seeding.np_random(seed)
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31.
seed2 = seeding.hash_seed(seed1 + 1) % 2**31
# Empirically, we need to seed before loading the ROM.
self.ale.setInt(b'random_seed', seed2)
def seed(self, seed: Optional[int] = None):
r"""Define random number generators for different aspects of
the environment.
"""
seed = seeding.hash_seed(seed, max_bytes=2)
self._rng = random.Random()
self._rng_read = random.Random()
self._rng_np = np.random.RandomState()
self._rng.seed(seed)
self._rng_read.seed(seed)
self._rng_np.seed(seed)
def get_experiment_environment(**args):
from utils import setup_mpi_gpus, setup_tensorflow_session
from baselines.common import set_global_seeds
from gym.utils.seeding import hash_seed
process_seed = args["seed"] + 1000 * MPI.COMM_WORLD.Get_rank()
process_seed = hash_seed(process_seed, max_bytes=4)
set_global_seeds(process_seed)
setup_mpi_gpus()
tf_context = setup_tensorflow_session()
return tf_context
def get_experiment_environment(**args):
from utils import setup_tensorflow_session
from baselines.common import set_global_seeds
from gym.utils.seeding import hash_seed
process_seed = args["seed"] + 1000 * 0
process_seed = hash_seed(process_seed, max_bytes=4)
set_global_seeds(process_seed)
logger_context = logger.scoped_configure(
dir=None, format_strs=['stdout', 'log', 'csv'])
tf_context = setup_tensorflow_session()
return logger_context, tf_context
def seed(self, seed=None):
self.np_random, seed1 = seeding.np_random(seed)
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31. * 2
seed2 = seeding.hash_seed(seed1 + 1)
# Empirically, we need to seed before loading the ROM (ignoring this for now in our case).
# return [seed1, seed2]
self.state = None
self.last_pop = 0
assert self.micro.map.getMapState().shape
def _seed(self, seed=None): self.ale.setFloat(b'repeat_action_probability', options.repeat_action_probability) from gym.utils import seeding self.np_random, seed1 = seeding.np_random(seed) # Derive a random seed. This gets passed as a uint, but gets # checked as an int elsewhere, so we need to keep it below # 2**31. seed2 = seeding.hash_seed(seed1 + 1) % 2 ** 31 # Empirically, we need to seed before loading the ROM. self.ale.setInt(b'random_seed', seed2) self.ale.loadROM(self.game_path) return [seed1, seed2]
def _seed(self, seed=None):
self.ale.setFloat(b'repeat_action_probability',
options.repeat_action_probability)
from gym.utils import seeding
self.np_random, seed1 = seeding.np_random(seed)
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31.
seed2 = seeding.hash_seed(seed1 + 1) % 2**31
# Empirically, we need to seed before loading the ROM.
self.ale.setInt(b'random_seed', seed2)
self.ale.loadROM(self.game_path)
return [seed1, seed2]
def _create_pipes(self):
# Creates named pipe for inter-process communication
self.pipe_name = seeding.hash_seed(None) % 2 ** 32
self.launch_vars['pipe_name'] = self.pipe_name
if not self.disable_out_pipe:
self.path_pipe_out = '%s-out.%d' % (self.path_pipe_prefix, self.pipe_name)
os.mkfifo(self.path_pipe_out)
# Launching a thread that will listen to incoming pipe
# Thread exits if self.is_exiting = 1 or pipe_in is closed
if not self.disable_in_pipe:
thread_incoming = Thread(target=self._listen_to_incoming_pipe, kwargs={'pipe_name': self.pipe_name})
thread_incoming.start()
def _seed(self, seed=None):
self.np_random, seed1 = seeding.np_random(seed)
# Derive a random seed.
seed2 = seeding.hash_seed(seed1 + 1) % 2**32
pachi_py.pachi_srand(seed2)
shape = pachi_py.CreateBoard(self.board_size).encode().shape
self.observation_space = spaces.Box(np.zeros(shape),
np.ones(shape),
np_random=self.np_random)
# One action for each board position, pass, and resign
self.action_space = spaces.Discrete(self.board_size**2 + 2,
np_random=self.np_random)
return [seed1, seed2]
def get_experiment_environment(**args):
process_seed = 1234 + 1000 * MPI.COMM_WORLD.Get_rank()
process_seed = hash_seed(process_seed, max_bytes=4)
set_global_seeds(1234)
setup_mpi_gpus()
logger_context = logger.scoped_configure(
dir='C:/Users/Elias/Desktop/savedunc/' + MODE + '_' +
datetime.now().strftime('%Y_%m_%d_%H_%M_%S'),
format_strs=['stdout', 'log', 'csv', 'tensorboard']
if MPI.COMM_WORLD.Get_rank() == 0 else ['log'])
tf_context = setup_tensorflow_session()
return logger_context, tf_context
def seed(self, mode=1, seed=None):
self.np_random, seed1 = seeding.np_random(seed)
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31.
seed2 = seeding.hash_seed(seed1 + 1) % 2**31
# Empirically, we need to seed before loading the ROM.
self.ale.setInt(b'random_seed', seed2)
self.ale.loadROM(self.game_path)
# set the mode, removing 1 from the number
# 1 is removed so the user can use the number from the manual
# mode 1 in the manual is mode 0 in the code
self.ale.setMode(mode - 1)
return [seed1, seed2]
def get_experiment_environment(**args):
from curiosity.utils import setup_mpi_gpus, setup_tensorflow_session
from baselines.common import set_global_seeds
from gym.utils.seeding import hash_seed
process_seed = args["seed"] + 1000 * MPI.COMM_WORLD.Get_rank()
process_seed = hash_seed(process_seed, max_bytes=4)
set_global_seeds(process_seed)
setup_mpi_gpus()
logger_context = logger.scoped_configure(dir=None,
format_strs=['stdout', 'log',
'csv'] if MPI.COMM_WORLD.Get_rank() == 0 else ['log'])
tf_context = setup_tensorflow_session()
return logger_context, tf_context
def seed(self, seed=None):
"""
This is totally the implementation in AtariEnv in openai/gym.
"""
self.np_random, seed1 = seeding.np_random(seed)
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31.
# Toybox takes a uint seed, but we're copying the ALE seed for reasons above.
# They're unclear who checks, so being safe here.
seed2 = seeding.hash_seed(seed1 + 1) % 2**31
self.toybox.set_seed(seed2)
# Start a new game to ensure that the seed gets used!.
self.toybox.new_game()
return [seed1, seed2]
def _seed(self, seed=None):
np_random, seed1 = seeding.np_random(seed)
# Derive a random seed.
seed2 = seeding.hash_seed(seed1 + 1) % 2**32
self.game.set_seed(seed2)
# 3 allowed actions [12, 13, 14] (must match .cfg file)
self.action_space = spaces.HighLow(np.matrix([[0, 1, 0]] * 3),
np_random=np_random)
self.observation_space = spaces.Box(low=0,
high=255,
shape=(self.screen_height,
self.screen_width, 3),
np_random=np_random)
return [seed1, seed2]
def _seed(self, seed=None):
np_random, seed1 = seeding.np_random(seed)
# Derive a random seed.
seed2 = seeding.hash_seed(seed1 + 1) % 2**32
self.game.set_seed(seed2)
# action indexes are [0, 9, 10, 12, 13, 14]
self.action_space = spaces.HighLow(np.matrix([[0, 1, 0]] * 6),
np_random=np_random)
self.observation_space = spaces.Box(low=0,
high=255,
shape=(self.screen_height,
self.screen_width, 3),
np_random=np_random)
return [seed1, seed2]
def _create_env_from_random_state(self):
#print('_create_env_from_random_state simglucose extended observation')
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31.
seed2 = seeding.hash_seed(self.np_random.randint(0, 1000)) % 2**31
seed3 = seeding.hash_seed(seed2 + 1) % 2**31
seed4 = seeding.hash_seed(seed3 + 1) % 2**31
hour = self.np_random.randint(low=0.0, high=24.0)
if self.saved_state is not None:
seed2=self.saved_state[0]
seed3=self.saved_state[1]
seed4=self.saved_state[2]
hour=self.saved_state[3]
print('Using state', seed2, seed3, seed4, hour)
start_time = datetime(2021, 1, 1, hour, 0, 0)
patient = T1DPatient.withName(self.patient_name, random_init_bg=True, seed=seed4)
sensor = CGMSensor.withName(self.SENSOR_HARDWARE, seed=seed2)
scenario = RandomScenario(start_time=start_time, seed=seed3)
pump = InsulinPump.withName(self.INSULIN_PUMP_HARDWARE)
env = _T1DSimEnvExtendedObs(patient, sensor, pump, scenario,n_samples=self._n_samples, append_action=self.append_action )
self.time_per_step=int(sensor.sample_time)
self.time_in_env=0
return env, seed2, seed3, seed4, hour
def _seed(self, seed=None):
self.np_random, seed1 = seeding.np_random(seed)
# Derive a random seed. This gets passed as a uint, but gets
# checked as an int elsewhere, so we need to keep it below
# 2**31.
seed2 = seeding.hash_seed(seed1 + 1) % 2**31
# Empirically, we need to seed before loading the ROM.
self.ale.setInt(b'random_seed', seed2)
self.ale.loadROM(self.game_path)
self._action_set = self.ale.getMinimalActionSet()
self.action_space = spaces.Discrete(len(self._action_set), np_random=self.np_random)
(screen_width,screen_height) = self.ale.getScreenDims()
if self._obs_type == 'ram':
self.observation_space = spaces.Box(low=np.zeros(128), high=np.zeros(128)+255, np_random=self.np_random)
elif self._obs_type == 'image':
self.observation_space = spaces.Box(low=0, high=255, shape=(screen_height, screen_width, 3), np_random=self.np_random)
else:
raise error.Error('Unrecognized observation type: {}'.format(self._obs_type))
return [seed1, seed2]
def _seed(self, seed=None):
seed = seed if seed else seeding.hash_seed(seed) % 2**32
random.seed(seed)
return [seed]
def _seed(self, seed=None):
self.np_random, seed1 = seeding.np_random(seed)
# Derive a random seed.
seed2 = seeding.hash_seed(seed1 + 1) % 2 ** 32
pachi_py.pachi_srand(seed2)
return [seed1, seed2]
def _seed(self, seed=None):
seed = seeding.hash_seed(seed) % 2**32
self.game.set_seed(seed)
return [seed]
def seed(self, seed=None):
self.np_random, seed1 = seeding.np_random(seed)
seed2 = seeding.hash_seed(seed1 + 1) % 2 ** 31
return [seed1, seed2]
def _seed(self, seed=None):
# Derive a random seed.
seed = seeding.hash_seed(seed) % 2**32
self.game.set_seed(seed)
return [seed]
def _seed(self, seed=None):
self.curr_seed = seeding.hash_seed(seed) % 2 ** 32
return [self.curr_seed]
