observations, _, _ = env.reset()
for obs in observations:
obs_queue.append(obs)
training = len(
memory) > WARM_STEPS # len(memory) = step，50000，先存储一些数据用于后续训练
state = env.make_state(obs_queue).to(device).float() # 丢掉第一个状态(~)
action = agent.run(state, training) # 选取动作a（以eps的概率随机选取动作a, 否则a由Q网络选取）
obs, reward, done = env.step(action) # 执行动作a, 获取奖励reward和下一个状态s’
obs_queue.append(obs) # 加入新的状态
memory.push(env.make_folded_state(obs_queue), action, reward,
done) # 储存（s,a,r）到经验池中
if step % POLICY_UPDATE == 0 and training:
agent.learn(memory, BATCH_SIZE) # 训练
if step % TARGET_UPDATE == 0: agent.sync() # 将策略网络中的权重同步到目标网络
if step % EVALUATE_FREQ == 0:
avg_reward, frames = env.evaluate(obs_queue, agent,
render=RENDER) # 让几个AI玩一下游戏，记录一下奖励
with open("rewards.txt", "a") as fp: # 文件记录奖励，用于后续画图
fp.write(f"{step//EVALUATE_FREQ:3d} {step:8d} {avg_reward:.1f}\n")
if RENDER: # No～
prefix = f"eval_{step//EVALUATE_FREQ:03d}"
os.mkdir(prefix)
for ind, frame in enumerate(frames):
with open(os.path.join(prefix, f"{ind:06d}.png"), "wb") as fp:
frame.save(fp, format="png")
agent.save(
os.path.join(SAVE_PREFIX,
f"model_{step//EVALUATE_FREQ:03d}")) # 保存模型
done = True
obs_queue.append(obs)
training = len(memory) > WARM_STEPS # 前面是热身
state = env.make_state(obs_queue).to(device).float()
action = agent.run(state, training)
obs, reward, done = env.step(action)
obs_queue.append(obs)
memory.push(env.make_folded_state(obs_queue), action, reward,
done) # 加入记忆池中
if step % POLICY_UPDATE == 0 and training:
agent.learn(memory, BATCH_SIZE) # 使用之前的记忆进行学习
if step % TARGET_UPDATE == 0: # 将policy的网络同步到Target中去
agent.sync()
if step % EVALUATE_FREQ == 0:
avg_reward, frames = env.evaluate(obs_queue, agent, render=RENDER)
with open("rewards.txt", "a") as fp:
fp.write(f"{step//EVALUATE_FREQ:3d} {step:8d} {avg_reward:.1f}\n")
if RENDER:
prefix = f"eval_{step//EVALUATE_FREQ:03d}"
os.mkdir(prefix)
for ind, frame in enumerate(frames):
with open(os.path.join(prefix, f"{ind:06d}.png"), "wb") as fp:
frame.save(fp, format="png")
agent.save(
os.path.join( # 保存当前模型
SAVE_PREFIX, f"model_{step//EVALUATE_FREQ:03d}"))
done = True
if RENDER:
prefix = f"eval_{step//EVALUATE_FREQ:03d}"
os.mkdir(prefix)
for ind, frame in enumerate(frames):
with open(os.path.join(prefix, f"{ind:06d}.png"),
"wb") as fp:
frame.save(fp, format="png")
#agent.save(os.path.join(SAVE_PREFIX, f"model_{step//EVALUATE_FREQ:03d}"))
done = True
avg_reward_arr.append(avg_reward)
progressive.set_description(
'current eval Avg Reward:{}'.format(avg_reward))
plt.plot(range(len(avg_reward_arr)),
avg_reward_arr,
co[version],
label=version)
np_arr = np.array(avg_reward_arr)
np.savetxt('./natural_result.txt'.format(version),
np_arr,
fmt='%f',
delimiter=',')
agent.save(os.path.join(SAVE_PREFIX, "model_{}.pth".format(version)))
if COUNT_SHADE:
action_queues.append(action_queue)
if COUNT_SHADE:
count_shade(action_queues, versions)
plt.legend()
plt.savefig('./result_running3.png')
