Thorsten Kurth
Thorsten Kurth
Hello @Weigaa, can you please hand us a simple reproducer for this problem. I lost track of what the latest code is, it would be good if you can just...
Looks good to me, I think this one is a no-brainer. Any objections?
What it does makes sense. Is this fix supposed to safeguard against initialized DM but requesting an uninitialized distributed group?
According to the documentation, the nodes in treelib should be ordered as they are appended. So if that is the case, I think we should be good.
Hi @luozero0, do you have more questions? Can we close this issue?
Hello, sorry for the delayed reply, I as off for the last few days. I will have a look tomorrow.
Ok, so here is the thing: SACMLP is different from a regular MLP in the sense that it produces a log-variance (or log-std to be more precise) prediction as well....
SAC is not the easiest algorithm to get to work properly, may I ask why you are not using something like PPO? Afaiu, PPO is widely used in the RL...
Using complex states is a tricky thing. In this case, I would write my own action policy to deal with the complex dof. Either using a neural operator architecture or...
If you want to use an action policy (and value function) which can handle more complex states, you need to write your own in pytorch, export it, and then use...