Honglong Tian
Honglong Tian
> Thanks for raising this, let me think about it, I guess we can do that in a vectorized way! @vmoens Yes, I also think the efficiency of these loops...
> There is a `break_when_any_done` argument that can be used to manipulate this. Would that solve it? @vmoens Unfortunately not, when `break_when_any_done=False`, all envs will be reseted when any one...
> In theory, rollout will only reset the envs that are done. You can check in the doc how this is done: we assign a "_reset" key corresponding to each...
> * To make the `rollout` step early we can design a special transform that keeps track of the dones, wdyt? You are right, and we can even do the...
@vmoens Cool! I have noticed that there is a issue listing some potential improvement (https://github.com/pytorch/rl/issues/916), I am really looking forward to it! And just like MPC, dreamer is also failed...
check https://github.com/polixir/emei plz