print "State: {}; Q: {}".format(state[0], Q)
total_score_l = []
sample_period = 5
num_episodes = 50000
starting_uncertainty = model.get_sigma_l()
sigma_average_dict = defaultdict(list)
components = ['W: first hidden', 'b: first hidden', 'W: second hidden', 'b: second hidden', \
'W: output','b: output']
for i_episode in range(num_episodes):
# Initialize the environment and state
env.reset()
state = Tensor(env.get_state()).unsqueeze(0)
score = 0
for t in xrange(500):
# Select and perform an action
if t % sample_period == 0:
w_sample = model.sample()
action = select_action(state)
reward, done = env.do_action(action[0, 0])
score += reward
reward = Tensor([reward])
# Observe new state
if not done:
next_state = Tensor(env.get_state()).unsqueeze(0)
else:
