Add support for tuple input on MultiBinary space #2023
Conversation
There was a problem hiding this comment.
I don't quite understand the desired behavior of this change. Do you want the samples to be multi-dimensional boolean tensors? That does not sound unreasonable, but why then there is a check of length != 1?
Arguably, contains should also be modified to check that the shape of the sample matches the shape of space (that's not implemented now, but I think it becomes more relevant if we allow multi-dimensional multibinary spaces)
Also, please update the docstring too to reflect the change
gym/spaces/multi_binary.py
Outdated
| @@ -32,7 +37,7 @@ def contains(self, x): | |||
|
|
|||
| def to_jsonable(self, sample_n): | |||
| return np.array(sample_n).tolist() | |||
|
|
|||
There was a problem hiding this comment.
please remove the spaces
|
@pzhokhov Thank you for all your suggestions. |
* Add support for tuple input on MultiBinary space * Change input of multibinary * Remove check of length != 1; Add shape check
When I'm using ray-project/ray, I've found out that when I'm using multi-dimensions space as the input n of
MultiBinary, problem occurs on init part. After debugging, it appeared that the dealing way of n is not suitable for multi-dimensions input.Here's the issue in ray-project/ray: #10024 on ray.