stanfordnmbl/osim-rl

Why the project = True obs_as_dict = True gives an array with different size (339 locally, 710 on crowdai evaluation)?

yhamidullah opened this issue · 6 comments

Hi,
I tried

  1. project = True obs_as_dict = True locally : gives me after flattening an array of shape (339,)
  2. project = False obs_as_dict = False locally : gives me after flattening and array of shape (710,)

But sending to evaluation at crowdai, knowing that it is setted project = True obs_as_dict = True, I expect to get the same shape (339,), but got (710,)

So I guess that in at crowdai it is setted project = False obs_as_dict = False
I trained my model with an input of the observation space so I can't do anything to reformat these 710 values to 339 to have the expected observation.

Is there a workaround?

Thanks

@yhamidullah : The evaluator uses this release , can you confirm if you face the issue even with this release ?

@kidzik : Should we make a new release with any new changes ? just a minor patch should do, as long as the release is also available on PyPI.

image

Yes I confirm that I use the release 3.0.3

@yhamidullah : Can you try making a submission again ?

@spMohanty Another submission is sent, and it's in evaluation queue actually, since 26 minutes.

Thanks very much!