Use num_batch_dims=0 to deal with observations containing scalars #231
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When a (nested) observation contain a scalar value on some axes
with dimension 1, batch_concat() will crash due to a shape mismatch
error (e.g., cannot concatenate tensors of shape [100, 1] and []).
This happens because batch_concat by default assumes the input
tensors are batched, but when computing the statistics
the batch dimensions (axis=0) already have been eliminated.
Note: This applies to SAC agent. PPO agent already has a correct implementation (comparison here).