for e in range(n_epochs):
# reset
frame = 0
loss = 0.0
Q_max = 0.0
total_reward = 0
env.reset()
state_t_1, reward_t, terminal = env.step(
np.random.choice(env.enable_actions))
while not terminal:
state_t = state_t_1
# execute action in environment
if steps > warmup:
action_t = agent.select_action(state_t)
else:
action_t = np.random.choice(env.enable_actions)
# observe environment
state_t_1, reward_t, terminal = env.step(action_t)
total_reward += reward_t
# store experience
agent.store_experience(state_t, action_t, reward_t, state_t_1,
terminal)
print(agent.tmp_q_values, np.argmax(agent.tmp_q_values),
agent.enable_actions.index(action_t))
# for log
frame += 1
