def pytest_runtest_call(item):
# seed = int(item.config.getoption('--pnl-seed'))
seed = 0
random.seed(seed)
np.random.seed(seed)
from psyneulink.core.globals.utilities import set_global_seed
set_global_seed(seed)
def pytest_runtest_call(item):
# seed = int(item.config.getoption('--pnl-seed'))
seed = 0
np.random.seed(seed)
set_global_seed(seed)
if 'pytorch' in item.keywords:
assert pytorch_available
torch.manual_seed(seed)
def __init__(self):
self.seed = int.from_bytes(os.urandom(4), byteorder="big")
from psyneulink.core.globals.utilities import set_global_seed
set_global_seed(self.seed)
np.random.seed(self.seed+1)
# Setup a Gym Forager environment for the game
self.gym_forager_env = ForagerEnv(obs_type='egocentric', incl_values=False, frameskip=2)
self.gym_forager_env.seed(self.seed+2)
# Setup an instance of the double DQN agent for determining optimal actions
self.ddqn_agent = DoubleDQNAgent(model_load_path=MODEL_PATH,
eval_mode=True,
save_frames=False,
render=RENDER,
env=self.gym_forager_env)
# Setup the PsyNeuLink composition
self._setup_composition()
from psyneulink import *
from double_dqn import DoubleDQNAgent
from gym_forager.envs.forager_env import ForagerEnv
parser = argparse.ArgumentParser()
parser.add_argument("--seed", type=int,
default=int.from_bytes(os.urandom(4), byteorder="big"),
help='Random seed, seed from os.urandom if unspecified.')
#args = parser.parse_args()
SEED = int.from_bytes(os.urandom(4), byteorder="big")
from psyneulink.core.globals.utilities import set_global_seed
set_global_seed(SEED)
np.random.seed(SEED+1)
# *********************************************************************************************************************
# *********************************************** CONSTANTS ***********************************************************
# *********************************************************************************************************************
# Runtime switches:
MPI_IMPLEMENTATION = True
RENDER = False
PNL_COMPILE = False
RUN = True
SHOW_GRAPH = False
MODEL_PATH = '../../../../double-dqn/models/trained_models/policy_net_trained_0.99_20190214-1651.pt'
# Switch for determining actual action taken in each step