# env.render()
# 4. Run agent on the state
action = mario.act(state)
# 5. Agent performs action
next_state, reward, done, info = env.step(action)
# 6. Remember
mario.cache(state, next_state, action, reward, done)
# 7. Learn
q, loss = mario.learn()
# 8. Logging
logger.log_step(reward, loss, q)
# 9. Update state
state = next_state
# 10. Check if end of game
if done or info['flag_get']:
break
logger.log_episode()
if e % print_e == 0:
logger.record(episode=e,
epsilon=mario.exploration_rate,
step=mario.curr_step)
episode_rewards.append(reward)
q, loss = doomguy.learn()
logger.log_step(reward, loss, q)
# If the game is finished
if done:
# the episode ends so no next state
next_state = np.zeros((84, 84), dtype=np.int)
next_state, stacked_frames = stack_frames(stacked_frames,
next_state, False)
# Get the total reward of the episode
total_reward = np.sum(episode_rewards)
break
else:
next_state, stacked_frames = get_next_state(False)
#memory.add((state, action, reward, next_state, done))
doomguy.cache(state, next_state, action, reward, done)
state = next_state
if step > max_steps:
total_reward = np.sum(episode_rewards)
break
logger.log_episode()
if episode % 20 == 0:
logger.record(episode=episode,
epsilon=doomguy.exploration_rate,
step=doomguy.curr_step)