b-vm
b-vm
@araffin have you been able to take a look at this yet? I am very curious what you think about it.
Cool. Let me know if you need any help running experiments/coding
My bad. Bug is fixed now!
Yes, it has only been implemented for Box action spaces so that might be it. I have not much time to work on this anymore. So feel free to do...
Thanks for the answers! I checked and you are right, the masking always yields the correct sized obs/act. I am still a bit confused on the purpose of padding in...
Thanks for your elaborate reply. It makes much more sense now. Although my models are training very well with SB3, I was still running into problems where it was taking...
> That's why you should try PPO with framestack first as we recommend in the doc. I have indeed read that, however I am expanding on prior research so I...
Fair points. I reran the test on BipedalWalker-v3 with the proper hyperparams (from sb3 zoo), and also ran a test on PendulumNoVel-v1. Here are the results: ### BipedalWalker-v3:  Orange...
Of course! Here it is: #118 Let me know if you want to see any changes.
> This is indeed a nice way to accelerate PPO LSTM. Cool! Glad to hear that. > you are probably never sampling the first sequence nor the last once (likely...