tutorial_BipedalWalker_v3.ipynb 跑不通

[xiezhipeng-git]

直接运行tutorial_BipedalWalker_v3.ipynb 报错

AttributeError Traceback (most recent call last) e:\study\machineStudy\project\matd3\ElegantRL\tutorial_BipedalWalker_v3.ipynb Cell 15 in <cell line: 1>() ----> 1 train_and_evaluate(args)

File e:\study\machineStudy\project\matd3\ElegantRL\elegantrl\train\run.py:91, in train_and_evaluate(args) 88 buffer = init_buffer(args, gpu_id) 89 evaluator = init_evaluator(args, gpu_id) ---> 91 agent.state = env.reset() 92 if args.if_off_policy: 93 trajectory = agent.explore_env(env, args.num_seed_steps * args.num_steps_per_episode)

AttributeError: 'NoneType' object has no attribute 'reset' 我试着使用args.env = build_env(env_func=args.env_func, env_args=args.env_args) 进行自己添加env参数。也会报其他错误。这个框架很好，模块都是分离的。要是有把各种算法环境都跑通。统一放进一个demo文件夹下就更好了。然后增加个mpe环境比较合适。多智能体的环境和demo太少

[shixun404]

@xiezhipeng-git Thanks for your reminder!

1. We are currently working on the tutorial_BipedalWalker_v3.ipynb. This bug could be solved quickly.
2. Besides, we will add MARL demos as soon as possible. Would you please help us test those demos after we update them?

[shixun404]

@xiezhipeng-git tutorial_BipedalWalker_v3.ipynb has been solved by Pull Request #214 . Please let us know if you have any feedback or suggestions :)

[xiezhipeng-git]

@xiezhipeng-git感谢您的提醒！

1. 我们目前正在开发_tutorial_BipedalWalker_v3.ipynb_。这个错误可以很快解决。
2. 此外，我们将尽快添加 MARL 演示。您能帮助我们在更新这些演示后对其进行测试吗？ @shixun404 1.更新后遇到了不一样的问题。下面是报错信息 2.我不一定一直会进行这个项目的测试啊，遇上了会试一下。有问题就报告

tutorial_BipedalWalker_v3.ipynb train_and_evaluate(args)

| Arguments Remove cwd: ./BipedalWalker-v3_PPO_0 Output exceeds the size limit. Open the full output data in a text editor

TypeError Traceback (most recent call last) e:\study\machineStudy\project\ElegantRL\ElegantRL\tutorial_BipedalWalker_v3.ipynb Cell 15 in <cell line: 1>() ----> 1 train_and_evaluate(args)

File e:\study\machineStudy\project\ElegantRL\ElegantRL\elegantrl\train\run.py:95, in train_and_evaluate(args) 92 env = build_env(args.env, args.env_func, args.env_args) 93 steps = 0 ---> 95 agent = init_agent(args, gpu_id, env) 96 buffer = init_buffer(args, gpu_id) 97 evaluator = init_evaluator(args, gpu_id)

File e:\study\machineStudy\project\ElegantRL\ElegantRL\elegantrl\train\run.py:24, in init_agent(args, gpu_id, env) 23 def init_agent(args: Arguments, gpu_id: int, env=None) -> AgentBase: ---> 24 agent = args.agent_class(args.net_dim, args.state_dim, args.action_dim, gpu_id=gpu_id, args=args) 25 agent.save_or_load_agent(args.cwd, if_save=False) 27 if env is not None:

File e:\study\machineStudy\project\ElegantRL\ElegantRL\elegantrl\agents\AgentPPO.py:40, in AgentPPO.init(self, net_dim, state_dim, action_dim, gpu_id, args) 38 self.cri_class = getattr(self, "cri_class", CriticPPO) 39 self.if_cri_target = getattr(args, "if_cri_target", False) ---> 40 AgentBase.init(self, net_dim, state_dim, action_dim, gpu_id, args) 42 self.ratio_clip = getattr( 43 args, "ratio_clip", 0.25 44 ) # could be 0.00 ~ 0.50 ratio.clamp(1 - clip, 1 + clip) ... --> 397 self.net = build_mlp_net(dims=[state_dim, *dims, action_dim]) 398 layer_init_with_orthogonal(self.net[-1], std=0.1) 400 self.action_std_log = nn.Parameter(torch.zeros((1, action_dim)), requires_grad=True)

TypeError: Value after * must be an iterable, not int

调试时看到dims 是16