def __init__(self, *args, **kwargs):
"""
[OPTIONAL]
Initialize the agent with the `test_case_id` (string), which might be important
if your agent is test case dependent.
For example, you might want to load the appropriate neural networks weight
in this method.
"""
test_case_id = kwargs.get("test_case_id")
"""
# Uncomment to help debugging
print(">>> __INIT__ >>>")
print("test_case_id:", test_case_id)
"""
self.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
env = construct_random_lane_env()
env.reset()
self.dqn = AtariDQN(env.observation_space.shape,
env.action_space.n).to(self.device)
self.dqn.load_state_dict(
torch.load(MODEL_NAME, map_location=lambda storage, loc: storage))
self.dqn.to(self.device)
self.dqn.eval()
def get_task():
tcs = [('t2_tmax50', 50), ('t2_tmax40', 40)]
return {
'time_limit':
600,
'testcases': [{
'id': tc,
'env': construct_random_lane_env(),
'runs': 300,
't_max': t_max
} for tc, t_max in tcs]
}
def get_task():
tcs = [("t2_tmax50", 50), ("t2_tmax40", 40)]
return {
"time_limit":
600,
"testcases": [{
"id": tc,
"env": construct_random_lane_env(),
"runs": 300,
"t_max": t_max
} for tc, t_max in tcs]
}
'''
# try: # server-compatible import statement
# from models import *
# except: pass
# try: # local-compatible import statement
# from .models import *
# except: pass
import torch
from mcts import *
from dqn import *
from env import *
from models import AtariDQN
import torch.optim as optim
env = construct_random_lane_env()
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
script_path = os.path.dirname(os.path.realpath(__file__))
model_path = os.path.join(script_path, 'model.pt')
train_steps = 10 #dqn training steps
T = 1
RANDOM_SEED = 1234
max_iterations = 100 #dagger iteration
numiters = 100 #mcts iterations
# Transition = collections.namedtuple('Transition', ('state', 'action', 'reward', 'next_state', 'done'))
def simulatePolicy(state, model, env):
'''
Policy followed in MCTS simulation for playout
'''