Albert Bou
Albert Bou
Where would the index key be set? in a Transform? or the environment itself would do it? I think the key could be a bool. named `padded` step or `skipped`...
The way I solved it in my case was to create a custom spec for every model I have and simply assign it to the model --> model.rnn_spec = spec....
yes, I think I will have some time, but maybe I need a bit of guidance. So the idea is: 1) Make sure that all training scripts log to wandb...
https://github.com/pytorch/rl/pull/1822
Having A3C would be very cool!
hi! thanks for the quick reply I installed it with `python -m pip install safe-mol` I am using Python 3.9.16
Interesting, so with your change it learns fine? I think what could be happening is that your sample tensors have more dimensions than the distribution parameters, and then some broadcasting...
Thanks a lot for the quick review!! I fixed the issues
even if if never catches the exception is try/except still problematic? for example if the code to compute the entropy is the following: ``` try: e = dist.entropy() except NotImplementedError:...
I am happy to help too. I agree that a SliceSampler tutorial would be nice. As I understand the current QDN tutorial stores UNPADDED fix-length trajectories that can contain done...