max_path_length = 100
ec = 100.0
k = 0.5
alpha = 0.85
tf.set_random_seed(0)
sess = tf.Session()
sess.__enter__()
# Instantiate the env
data = joblib.load("../CartPole/ControlPolicy/itr_5.pkl")
sut = data['policy']
reward_function = ASTRewardS()
simulator = CartpoleSimulator(sut=sut,
max_path_length=max_path_length,
use_seed=False,
nd=1)
env = TfEnv(
ASTEnv(
open_loop=open_loop,
simulator=simulator,
fixed_init_state=True,
s_0=[0.0, 0.0, 0.0 * math.pi / 180, 0.0],
reward_function=reward_function,
))
# Training
with open(osp.join(args.log_dir, 'total_result.csv'), mode='w') as csv_file:
fieldnames = ['step_count']
for i in range(top_k):
fieldnames.append('reward ' + str(i))
k = args.k
alpha = args.alpha
rsg_length = 2
tf.set_random_seed(0)
sess = tf.Session()
sess.__enter__()
# Instantiate the env
data = joblib.load("../CartPole/ControlPolicy/itr_" + str(args.sut_itr) +
".pkl")
sut = data['policy']
reward_function = ASTRewardS()
simulator = CartpoleSimulator(sut=sut,
max_path_length=max_path_length,
use_seed=True,
nd=args.nd)
env = TfEnv(
ASTEnv(
open_loop=open_loop,
simulator=simulator,
fixed_init_state=True,
s_0=[0.0, 0.0, 0.0 * math.pi / 180, 0.0],
reward_function=reward_function,
))
# Training
with open(osp.join(args.log_dir, 'total_result.csv'), mode='w') as csv_file:
fieldnames = ['step_count']
for i in range(top_k):
fieldnames.append('reward ' + str(i))
seed = 0
top_k = 10
max_path_length = 100
top_paths = BPQ.BoundedPriorityQueue(top_k)
np.random.seed(seed)
tf.set_random_seed(seed)
with tf.Session() as sess:
# Create env
data = joblib.load("../CartPole/ControlPolicy/itr_5.pkl")
sut = data['policy']
reward_function = ASTRewardS()
simulator = CartpoleSimulator(sut=sut, max_path_length=100, use_seed=False)
env = ASTEnv(open_loop=False,
simulator=simulator,
fixed_init_state=True,
s_0=[0.0, 0.0, 0.0 * math.pi / 180, 0.0],
reward_function=reward_function,
)
env = TfEnv(env)
# Create policy
policy = DeterministicMLPPolicy(
name='ast_agent',
env_spec=env.spec,
hidden_sizes=(64, 32),
output_nonlinearity=tf.nn.tanh,
)
top_k = 10
top_paths = BPQ.BoundedPriorityQueue(top_k)
np.random.seed(seed)
tf.set_random_seed(seed)
with tf.Session() as sess:
# Create env
control_policy_path = "../CartPole/Data/Train/itr_5.pkl"
# control_policy_path = "../CartPole/control_policy.pkl"
print(control_policy_path)
data = joblib.load(control_policy_path)
sut = data['policy']
reward_function = ASTRewardS()
simulator = CartpoleSimulator(sut=sut, max_path_length=100, use_seed=True, nd=1)
env = ASTEnv(open_loop=False,
simulator=simulator,
fixed_init_state=True,
s_0=[0.0, 0.0, 0.0 * math.pi / 180, 0.0],
reward_function=reward_function,
)
env = TfEnv(env)
algo = MCTSRS(
env=env,
stress_test_num=2,
max_path_length=100,
ec=10.0,
n_itr=100,
k=0.5,