Antonin RAFFIN

> I don't find where specifically such a warning is raised, but there's definitely a codepath that expects to find the compute_reward() method enforced by the GoalEnv interface. well, the...

> I wanted to check if there's something to be done in SB3 regarding the removal of GoalEnv from gym, in which case we should probably address it in this...

Hello, i definitely agree with this feature (it was in my internal roadmap). i've tried a bit in the past but that was unsuccessful because jit support was limited and...

> @araffin I would like to help getting this PR merged. Are you or anyone else working on the missing points? If not, could I work on PRs to the...

> 2\. Zoo - Tests are working with sb3 at this fork, what test did you run for the zoo? by running `make pytest` with this branch, I already get...

Hello, > This might lead to undesirable effects for users relying on the last values in the callback. I'm curious to know more about your use-case. Callback may not be...

that's what i thought, you should fork SB3 for your usecase, otherwise you would compute the advantage twice and will have hacks with side effects here and there if you...

If you use callback for accessing both model and env, this is the same as forking, but more hacky... > Since I use SubProcEnv for parallel rollouts, it is not...

Hello, Thanks for the suggestion, this was introduced in https://github.com/pytorch/pytorch/pull/19785, so we need to check which PyTorch version is needed (hopefully 1.4 is enough). I would like also to make...

>So we added basic support for multiagent environments to stable baselines with a third party wrapper. I wrote a small tutorial on how to use it targeted at beginners in...