[Question] cv2 error on using expert_dataset.py (follow-up from DeprecationWarning: DataPipeline.sarsd_iter)

[prabhasak]

Hello. There is a cv2 error on running any baseline code from general/chainerrl/baselines (after fixing this issue). Here's the traceback on running the GAIL.py baseline:

Traceback (most recent call last):
  File "gail.py", line 421, in <module>
    main()
  File "gail.py", line 146, in main
    _main(args)
  File "gail.py", line 291, in _main
    shuffle=False)
  File "C:\MineRL\baselines\general\chainerrl\baselines\expert_dataset.py", line 110, in __init__
    self._convert(original_dataset)
  File "C:\MineRL\baselines\general\chainerrl\baselines\expert_dataset.py", line 120, in _convert
    obs = self.observation_converter(orig_obs)
  File "C:\MineRL\baselines\general\chainerrl\baselines\observation_wrappers.py", line 24, in converter
    cv2.cvtColor(obs, cv2.COLOR_RGB2GRAY), axis=-1)
cv2.error: OpenCV(4.3.0) c:\projects\opencv-python\opencv\modules\imgproc\src\color.simd_helpers.hpp:92: error: (-2:Unspecified error) in function '__cdecl cv::impl::`anonymous-namespace'::CvtHelper<struct cv::impl::`anonymous namespace'::Set<3,4,-1>,struct cv::impl::A0xe227985e::Set<1,-1,-1>,struct cv::impl::A0xe227985e::Set<0,2,5>,2>::CvtHelper(const class cv::_InputArray &,const class cv::_OutputArray &,int)'
> Invalid number of channels in input image:
>     'VScn::contains(scn)'
> where
>     'scn' is 1

Exception in thread QueueManagerThread:
Traceback (most recent call last):
  File "C:\Users\prabh\Miniconda3\envs\minerlpk\lib\threading.py", line 917, in _bootstrap_inner
    self.run()
  File "C:\Users\prabh\Miniconda3\envs\minerlpk\lib\threading.py", line 865, in run
    self._target(*self._args, **self._kwargs)
  File "C:\Users\prabh\Miniconda3\envs\minerlpk\lib\concurrent\futures\process.py", line 354, in _queue_management_worker
    ready = wait(readers + worker_sentinels)
  File "C:\Users\prabh\Miniconda3\envs\minerlpk\lib\multiprocessing\connection.py", line 872, in wait
    ov.cancel()
OSError: [WinError 6] The handle is invalid

Failed to delete the temporary minecraft directory.

1. On tracing the issue, turns out what's fed into self.observation_converter() and self.action_converter() variables has shape (1, 32, 64, 64, 3) instead of (32, 64, 64, 3) (introduced from the new DataPipeline.batch_iter(**1**, 32, 1))

2. So, in both observation_wrappers.py and action_wrappers.py we need to ignore the first dim (use dict[keys][0] instead of dict[keys]) and append the dim on return (return np.expand_dims(list, 0) instead of return list). This seems to fix the problem

Note: dict = observation, action and list = ret, val in the two files, respectively

PS: If the spaces weren't dict, we could fix this by just initializing the converters as lambda x: x[0] instead of lambda x: x :v

System Info

• Windows 10
• Python 3.7.0
• Tensorflow 2.2.0
• minerl 0.3.5