运行DI-Zoo里面写好的config报错

[xujinming01]

• I have marked all applicable categories:
  • exception-raising bug
  • RL algorithm bug
  • system worker bug
  • system utils bug
  • code design/refactor
  • documentation request
  • new feature request
• I have visited the readme and doc
• I have searched through the issue tracker and pr tracker
• I have mentioned version numbers, operating system and environment, where applicable:

  import ding, torch, sys
  print(ding.__version__, torch.__version__, sys.version, sys.platform)

v0.4.7 1.12.1+cpu 3.9.12 (main, Apr  4 2022, 05:22:27) [MSC v.1916 64 bit (AMD64)] win32

直接在Pycharm里面运行DI-engine\dizoo\gym_hybrid\config\gym_hybrid_pdqn_config.py报错：

Traceback (most recent call last):
  File "C:\Users\arche\code\repository\DI-engine\dizoo\gym_hybrid\config\gym_hybrid_pdqn_config.py", line 76, in <module>
    serial_pipeline([main_config, create_config], seed=0, max_env_step=int(1e7))
  File "C:\Users\arche\code\repository\DI-engine\ding\entry\serial_entry.py", line 97, in serial_pipeline
    stop, eval_info = evaluator.eval(learner.save_checkpoint, learner.train_iter, collector.envstep)
  File "C:\Users\arche\code\repository\DI-engine\ding\worker\collector\interaction_serial_evaluator.py", line 314, in eval
    return_info = to_item(return_info)
  File "C:\Users\arche\code\repository\DI-engine\ding\torch_utils\data_helper.py", line 301, in to_item
    return [to_item(d) for d in data]
  File "C:\Users\arche\code\repository\DI-engine\ding\torch_utils\data_helper.py", line 301, in <listcomp>
    return [to_item(d) for d in data]
  File "C:\Users\arche\code\repository\DI-engine\ding\torch_utils\data_helper.py", line 303, in to_item
    return {k: to_item(v) for k, v in data.items()}
  File "C:\Users\arche\code\repository\DI-engine\ding\torch_utils\data_helper.py", line 303, in <dictcomp>
    return {k: to_item(v) for k, v in data.items()}
  File "C:\Users\arche\code\repository\DI-engine\ding\torch_utils\data_helper.py", line 299, in to_item
    return data.item()
ValueError: only one element tensors can be converted to Python scalars

调试查看了一下 return_info 的信息：

[{'eval_episode_return': tensor([-1.2000]), 'action_args_mask': tensor([[1., 0.],
        [0., 1.],
        [0., 0.]])}, {'eval_episode_return': tensor([-1.2000]), 'action_args_mask': tensor([[1., 0.],
        [0., 1.],
        [0., 0.]])}, {'eval_episode_return': tensor([-1.2000]), 'action_args_mask': tensor([[1., 0.],
        [0., 1.],
        [0., 0.]])}, {'eval_episode_return': tensor([-1.2000]), 'action_args_mask': tensor([[1., 0.],
        [0., 1.],
        [0., 0.]])}, {'eval_episode_return': tensor([-1.2000]), 'action_args_mask': tensor([[1., 0.],
        [0., 1.],
        [0., 0.]])}]

是不是 to_item() 不支持多维的 tensor

[PaParaZz1]

这个问题已经在最新的 main 分支 commit 中修复了，你可以看下你那边运行环境是否还存在该问题。

[xujinming01]

Ok, thanks.