# batch train
total_reward = 0
env.reset()
action = env.action_space.sample()
state, reward, done, _ = env.step(action)
for _ in range(1000):
# training
states, actions, rewards, next_states = memory.sample(20)
next_actions = actor.get_actions(next_states)
next_qs = critic.get_qs(next_states, next_actions)
loss, q = critic.train(states, actions, rewards, next_qs)
action_gradients = critic.get_action_gradients(states, actions)
actor.train(states, action_gradients[0])
env.render()
action = actor.get_action_for_train(state, ep)
next_state, reward, done, _ = env.step(action)
memory.add((state, action, reward, next_state))
# print(state, action, reward, next_state)
total_reward += reward
# print(action, reward, total_reward)
state = next_state
if done:
break
# if ep % 10 == 0:
# critic.update_network_params()
logging.info('Episode: {}'.format(ep) +
' Total Reward: {:.4f}'.format(total_reward) +
' Q: {:.4f}'.format(np.max(q)) +
' loss: {:.4f}'.format(loss))
