M. Ernestus

For others that run into the same issue: A temporary workaround for me is this: ```python def numpy_transform(batch): return {key: np.asarray(val) for key, val in batch.items()} dataset = dataset.with_transform(numpy_transform) ```

Came here just to report the same bug. So let me just add this screenshot to emphasize the seriousness of this bug: 

All right. Will do. Thanks for the huge initial chunk of work! May I ask what other framework you ended up using @gabriel-v?

> Did a custom reinterpretation of this repo's DAG definition API using async python and scylladb, seeing some great throughput out of it - will leave a comment here when...

Hey @lavine-lmu, thanks for bringing this up. We plan on adding a tutorial but we first focus on doing a v1.0 release. In the meantime we are happy about any...

Looks like this PR was abandoned. I took the liberty to implement what you agreed upon.

@AdamGleave would you review this?

> Are no changes needed in `train_imitation.py`? No should have mentioned it. In `train_imitation.py` it is already cast to `int`.

The rl-zoo uses a `VecNormalize` environment wrapper to do normalize observations as well as rewards. Whenever a model is loaded using the zoo scripts (such as to continue training or...

I found a better solution: add the feature to use python modules as configs instead of yaml files. That lets us solve all of this without any hacks like factory...